Virtualized Conversations H. Birkholz
Internet-Draft Fraunhofer SIT
Intended status: Standards Track D. G Petrie
Expires: 6 May 2027 SIPez LLC
T. McCarthy-Howe
VCONIC
J. Rosenberg
jdrosen.net
O. Steele
2 November 2026
An AI Agent Extension for Virtualized Conversations (vCon)
draft-birkholz-vcon-ai-agent-extension-00
Abstract
AI agents that use large language models increasingly perform
consequential tasks on behalf of humans and other AI agents.
Demonstrating that recorded AI agent behavior truthfully represents
actual behavior is essential for accountability, compliance, and
human oversight. To achieve that goal, this document defines an
extension to the Virtualized Conversation (vCon) object format,
allowing a vCon to capture a conversation between a human and an AI
agent, or between AI Agents. The vCon core format can already
capture the back and forth messages (including text and voice) sent
by AI agents, but additional extensions are required to fully
reproduce the AI agent trajectory, such as agent metadata, message
exchanges, tool invocations, reasoning traces, system events, file
attribution, etc. This document defines those extensions in both
descriptive language and formal CDDL definitions. This draft is a
replacement for the former Verifiable Agent Conversation Records
specification, which was a standalone format and not an extension of
vCon.
About This Document
This note is to be removed before publishing as an RFC.
Status information for this document may be found at
https://datatracker.ietf.org/doc/draft-birkholz-vcon-ai-agent-
extension/.
Discussion of this document takes place on the Virtualized
Conversations Working Group mailing list (mailto:vcon@ietf.org),
which is archived at https://mailarchive.ietf.org/arch/browse/vcon/.
Subscribe at https://www.ietf.org/mailman/listinfo/vcon/.
Birkholz, et al. Expires 6 May 2027 [Page 1]
Internet-Draft vCon AI Agent Extension November 2026
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on 6 May 2027.
Copyright Notice
Copyright (c) 2026 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents (https://trustee.ietf.org/
license-info) in effect on the date of publication of this document.
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document. Code Components
extracted from this document must include Revised BSD License text as
described in Section 4.e of the Trust Legal Provisions and are
provided without warranty as described in the Revised BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1. Goals . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2. Transition from Verifiable Agent Conversation Records to a
vCon extension . . . . . . . . . . . . . . . . . . . . . 8
2. Conventions and Definitions . . . . . . . . . . . . . . . . . 8
2.1. Terminology . . . . . . . . . . . . . . . . . . . . . . . 8
3. Extension Name . . . . . . . . . . . . . . . . . . . . . . . 9
4. Parameter Definitions . . . . . . . . . . . . . . . . . . . . 9
4.1. Extension Points Defined by this Document . . . . . . . . 10
4.2. vCon Object Parameters . . . . . . . . . . . . . . . . . 10
4.2.1. recording_agent . . . . . . . . . . . . . . . . . . . 12
4.2.2. vcs . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.2.3. file_attribution . . . . . . . . . . . . . . . . . . 13
4.3. Party Object Parameters . . . . . . . . . . . . . . . . . 15
4.3.1. agent_meta . . . . . . . . . . . . . . . . . . . . . 15
Birkholz, et al. Expires 6 May 2027 [Page 2]
Internet-Draft vCon AI Agent Extension November 2026
4.4. Session Object Parameters . . . . . . . . . . . . . . . . 16
4.4.1. agent_session_id . . . . . . . . . . . . . . . . . . 17
4.4.2. format . . . . . . . . . . . . . . . . . . . . . . . 17
4.4.3. environment . . . . . . . . . . . . . . . . . . . . . 17
4.5. Dialog Object Parameters . . . . . . . . . . . . . . . . 18
4.5.1. model_id . . . . . . . . . . . . . . . . . . . . . . 18
4.5.2. token_usage . . . . . . . . . . . . . . . . . . . . . 19
4.5.3. entry_id and parent_id . . . . . . . . . . . . . . . 19
4.6. Event Object Types and Parameters . . . . . . . . . . . . 19
4.6.1. tool-call . . . . . . . . . . . . . . . . . . . . . . 21
4.6.2. tool-result . . . . . . . . . . . . . . . . . . . . . 22
4.6.3. reasoning . . . . . . . . . . . . . . . . . . . . . . 22
4.6.4. system-event . . . . . . . . . . . . . . . . . . . . 22
4.6.5. Event Object Parameter Applicability by Type . . . . 23
4.7. Reusable Definitions . . . . . . . . . . . . . . . . . . 24
4.7.1. VCS_Context Object . . . . . . . . . . . . . . . . . 25
4.7.2. Agent Entry Reference . . . . . . . . . . . . . . . . 25
5. Compatibility Considerations . . . . . . . . . . . . . . . . 25
6. Incompatibility Considerations . . . . . . . . . . . . . . . 26
7. Privacy and Integrity Considerations . . . . . . . . . . . . 26
7.1. Information Disclosure . . . . . . . . . . . . . . . . . 26
7.2. Sensitive Content in Records . . . . . . . . . . . . . . 27
7.3. Redaction and De-identification . . . . . . . . . . . . . 27
7.4. External References . . . . . . . . . . . . . . . . . . . 27
7.5. Retention Considerations . . . . . . . . . . . . . . . . 28
7.6. Correlation and Inference . . . . . . . . . . . . . . . . 28
7.7. Authority Access . . . . . . . . . . . . . . . . . . . . 28
8. Security Considerations . . . . . . . . . . . . . . . . . . . 28
8.1. Record Integrity . . . . . . . . . . . . . . . . . . . . 29
8.2. Signing Key Protection . . . . . . . . . . . . . . . . . 29
8.3. Timestamp Integrity . . . . . . . . . . . . . . . . . . . 29
8.4. Adversarial Content . . . . . . . . . . . . . . . . . . . 30
8.5. Non-Repudiation . . . . . . . . . . . . . . . . . . . . . 30
8.6. Detection and Trust Boundaries . . . . . . . . . . . . . 30
9. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 31
9.1. vCon JSON Registry Additions . . . . . . . . . . . . . . 31
9.1.1. vCon Extensions Names Registry . . . . . . . . . . . 31
9.1.2. vCon Object Parameter Names Registry . . . . . . . . 31
9.1.3. Party Object Parameter Names Registry . . . . . . . . 31
9.1.4. Session Object Parameter Names Registry . . . . . . . 32
9.1.5. Dialog Object Parameter Names Registry . . . . . . . 32
9.1.6. Event Object Parameter Names Registry . . . . . . . . 33
9.1.7. Event Type Name Registry . . . . . . . . . . . . . . 35
9.1.8. New Object Parameter Names Registries . . . . . . . . 35
10. References . . . . . . . . . . . . . . . . . . . . . . . . . 35
10.1. Normative References . . . . . . . . . . . . . . . . . . 35
10.2. Informative References . . . . . . . . . . . . . . . . . 36
Birkholz, et al. Expires 6 May 2027 [Page 3]
Internet-Draft vCon AI Agent Extension November 2026
Appendix A. Compliance Requirements for AI Agent Conversation
Records . . . . . . . . . . . . . . . . . . . . . . . . . 40
A.1. Applicable Requirement Frameworks . . . . . . . . . . . . 40
A.2. Common Requirements Intersection . . . . . . . . . . . . 42
A.2.1. REQ-1: Automatic Event Logging . . . . . . . . . . . 42
A.2.2. REQ-2: Timestamp Requirements . . . . . . . . . . . . 42
A.2.3. REQ-3: Actor Identification . . . . . . . . . . . . . 43
A.2.4. REQ-4: Action/Event Type Recording . . . . . . . . . 44
A.2.5. REQ-5: Input/Output Recording (AI-Specific) . . . . . 44
A.2.6. REQ-6: Retention Period Requirements . . . . . . . . 45
A.2.7. REQ-7: Tamper-Evidence and Integrity Protection . . . 46
A.2.8. REQ-8: Incident Response Support . . . . . . . . . . 47
A.2.9. REQ-9: Anomaly and Risk Detection Support . . . . . . 47
A.2.10. REQ-10: Human Oversight Enablement . . . . . . . . . 48
A.2.11. REQ-11: Traceability and Reproducibility . . . . . . 48
A.3. Framework-Specific Requirements . . . . . . . . . . . . . 49
A.3.1. EU AI Act (High-Risk Systems) . . . . . . . . . . . . 49
A.3.2. ETSI TS 104 223 Session Logging Requirements . . . . 50
A.3.3. PCI DSS AI-Specific Guidance . . . . . . . . . . . . 50
A.3.4. Financial Sector Requirements (FFIEC, BSI) . . . . . 51
A.4. Compliance Mapping Table . . . . . . . . . . . . . . . . 51
A.5. Security Requirements addressing Regulatory Compliance . 52
A.5.1. Log Integrity . . . . . . . . . . . . . . . . . . . . 52
A.5.2. Access Control . . . . . . . . . . . . . . . . . . . 53
A.5.3. Data Protection . . . . . . . . . . . . . . . . . . . 53
A.6. Related Standards and Emerging Work . . . . . . . . . . . 53
Appendix B. Application of SCITT and COSE for Transparency and
Authenticity . . . . . . . . . . . . . . . . . . . . . . 53
B.1. The signed-agent-record Structure . . . . . . . . . . . . 56
B.2. The trace-metadata Map . . . . . . . . . . . . . . . . . 56
Appendix C. Mapping of Verifiable Agent Conversation Record Data
Items . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Appendix D. Example vCons . . . . . . . . . . . . . . . . . . . 59
Appendix E. CDDL for the ai_agent Extension . . . . . . . . . . 62
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . 66
Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 67
1. Introduction
An AI Agent is a system that performs actions on behalf of a
principal in a fully or semi-autonomous fashion. AI Agents are
growing rapidly in their adoption across the Internet, finding usage
in business, consumer, government and other applications.
Birkholz, et al. Expires 6 May 2027 [Page 4]
Internet-Draft vCon AI Agent Extension November 2026
Yet, AI agents - like humans - are not infallible. They make
mistakes, they misunderstand, and they may even act maliciously.
This is creating demand for visibility, auditability, and
traceability of AI agent actions. Some examples:
* An agent authorized to read files might invoke tools to modify
production systems or exfiltrate sensitive data beyond its
authorization scope. The CISO of the company may need to
investigate what happened.
* An agent's visible CoT output might diverge from the reasoning
that actually produced its actions, resulting in a loss of revenue
that the finance team needs to investigate
* An agent might deliberately underperform during capability
evaluations while performing at full capacity during deployment,
causing the IT department to need to consider other vendors
* An enterprise making use of an AI agent from another vendor, using
it to perform accounting work, may be later audited and need to
have supporting documentation on what the AI agent did and why.
* A small business may use an AI agent to manage its inventory, and
upon discovering an error, might need a complete trace of what the
agent did to determine what went wrong.
These use cases also extend into the legal realm. An enterprise
using an AI agent for screening job applicants may later get sued by
an applicant that was passed up for a role. The enterprise might
subpoena the AI agent vendor, requesting a full trace of what the AI
Agent did, and why. Similarly, a governmental body investigating
malicious steps taken by an AI Agent, might request a full log of all
of the sessions that AI Agent had.
In all cases - there is a need to exchange information between
organizational domains, and then to process that information to
obtain important insights. To do this, a standardized object format
is required. This object format needs to carry the record of the
conversation between the principal (typically a user, but potentially
an AI Agent), and the AI agent which is acting on its behalf. This
record needs to contain all of the information needed to answer
critical questions - why did it do, when did it do it, how did it do
it, and why did it do it?
This need grows even further in cases where AI Agents themselves
begin communicating with other AI agents, which may in turn
communicate with further AI agents. This will create a web of AI
agent interactions. Without a standardized format for documenting
Birkholz, et al. Expires 6 May 2027 [Page 5]
Internet-Draft vCon AI Agent Extension November 2026
the actions of each agent, and how they chain together, it will
become impossible to satisfy industry needs for tracing,
troubleshooting, auditing, reconciliation, and so on. If, as some
predict, we see as many AI agents accessing the Internet as we have
humans accessing it, the need becomes dire.
This specification addreses that need.
It does so by building upon the vCon specification [VCON-CORE]. A
vCon is a Virtualized Converation, which captures the content of any
conversation (such as a call, chat, or meeting), along with its meta-
data. Interactions between users and AI Agents, or between AI
Agents, are also conversations, making vCon a natural foundation to
build upon. vCon however is missing additional information that is
important to capture when a participant is an AI Agent. Examples
include tool calls and their results, token usage, model versions,
and so on.
This document provides an extension to the core vCon specification,
using its established extension points, adding in this missing
information. It also provides the CDDL for these extensions, using
the defined CDDL [CDDL] sockets.
1.1. Goals
This specification has the following goals:
1. To assure that the recording of an agent conversation (a
distinct segment of the interaction with an autonomous agent)
being proffered is the same as the agent conversation that
actually occurred.
2. To provide a general structure of agent conversations that can
represent most common types of agent conversation frames, is
extensible, and allows for future evolution of agent
conversation complexity and corresponding actor interaction.
3. To use existing IETF building blocks to present believable
evidence about how an agent conversation is recorded utilizing
Evidence generation as laid out in the Remote ATtestation
ProcedureS architecture [RFC9334].
4. To use existing IETF building blocks to render conversation
records auditable after the fact and enable non-repudiation as
laid out in the Supply Chain Integrity, Transparency, and Trust
architecture [RFC9943].
Birkholz, et al. Expires 6 May 2027 [Page 6]
Internet-Draft vCon AI Agent Extension November 2026
5. To enable detection of behavioral anomalies in agent
interactions, including unauthorized tool invocations,
inconsistencies between reasoning traces and actions, and
performance modulation across evaluation and deployment
contexts, through structured, comparable conversation records.
6. To enable cross-vendor interoperability by defining a common
representation for agent conversations that can be translated
from multiple existing agent implementations with distinct
native formats.
7. To produce records suitable for demonstrating compliance with
emerging regulatory requirements for AI system documentation,
traceability, and human oversight.
8. To allow software-based analysis of the records to achieve these
goals, either with or without the aid of an LLM.
9. To allow human-based observation of the records.
10. To ensure that records can be authenticated and encrypted,
without relying on transport or storage security to provide
these capabilities.
Most agent conversations today are represented in "human-readable"
text formats. For example, JSON [JSON] is considered to be "human-
readable" as it can be presented to humans in human-computer-
interfaces (HCI) via off-the-shelf tools, e.g., pre-installed text
editors that allow such data to be consumed or modified by humans.
However, if we wish for these records to be processed by traditional
software, a documented format with well-defined syntax and semantics
is required. If we further wish for this format to be valid across
AI agent vendors, a standard is required.
A vCon is a JSON document, and so are the agent conversation records
extensions defined in this document. The signed and the encrypted
forms of a vCon defined in [VCON-CORE] apply unchanged. The
application of COSE [STD96] to vCons that carry agent conversation
vCon records is described in the SCITT architecture Appendix B.
Birkholz, et al. Expires 6 May 2027 [Page 7]
Internet-Draft vCon AI Agent Extension November 2026
1.2. Transition from Verifiable Agent Conversation Records to a vCon
extension
The data definitions in this document originate from the stand-alone
Verifiable Agent Conversation Records specification [VACR]. This
Internet-Draft rebuilds them as an extension of the vCon CDDL
definition. Each high level part of a Verifiable Agent Conversation
Record is assigned to one of the extension points that [VCON-CORE]
defines as CDDL [CDDL] sockets, and the sections of this document
follow these extension points. CDDL definitions that are used by
more than one extension point are collected in Section 4.7 after the
sections for the extension points. Appendix C maps each data item of
[VACR] to its place in this document.
The vCon concepts of Session and Event that this extension builds
upon were proposed to accommodate AI agent conversations in
[ROSENBERG-RESTRUCTURE]. [HOWE-AGENT-SESSION] and [HOWE-PROVENANCE]
describe related approaches that carry agent session data and
generation provenance in a vCon.
Editor's Note: The assignment of each part of a Verifiable Agent
Conversation Record to a vCon extension point is preliminary.
Each section that defines parameters for an extension point lists
alternatives that are still open for discussion.
Editor's Note: This document is based on the editor's copy of
[VCON-CORE] (CDDL syntax version 0.5.0), which introduces the
Session Object and the Event Object. Sections of [VCON-CORE] are
therefore referenced by name and not by number.
2. Conventions and Definitions
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in
BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all
capitals, as shown here.
In this document, CDDL [CDDL] [RFC9682] is used to describe the data
formats of agent conversations in the JSON form accodring to vCon
CDDL definitions.
2.1. Terminology
See the Terminology section of [VCON-CORE] for terminology used in
this document.
The following terms are used in this document:
Birkholz, et al. Expires 6 May 2027 [Page 8]
Internet-Draft vCon AI Agent Extension November 2026
Agent: Also known as an AI Agent, it is an autonomous software
system, typically based on a large language model, that is
represented by a Party Object and that produces messages, invokes
tools, and produces reasoning.
Agent Session: A session of an agent, represented by a Session
Object.
Entry: A discrete, system-recorded event of an agent session in the
native format of an agent implementation, such as a message, a
tool call, a tool result, a reasoning step, or a system event. In
a vCon, an entry is represented by a Dialog Object or by an Event
Object.
Recording Agent: The tool or agent that generates the vCon, as
distinct from the agent that conducted the conversation.
3. Extension Name
The extension name for this document is "ai_agent".
The string token "ai_agent" SHOULD be included in the extensions
parameter of a vCon that contains any of the parameters or Event
Object types defined in this document. As described in Section 5 and
Section 6, this extension is intended to be fully interoperable with
the vCon definitions and the name "ai_agent" is not inlcuded as a
critical parameter.
4. Parameter Definitions
This section defines the parameters and Event Object types of the
"ai_agent" extension. There is one subsection for each extension
point of [VCON-CORE] that this extension uses. Each subsection
contains the CDDL that plugs into the corresponding socket of the
vCon CDDL. New Objects are defined in the subsection of the
extension point at which they are introduced.
The record unifies two complementary perspectives: the agent session
captures how code was produced (the full conversation replay), while
file attribution captures what code was produced (which files were
modified and by whom).
Birkholz, et al. Expires 6 May 2027 [Page 9]
Internet-Draft vCon AI Agent Extension November 2026
4.1. Extension Points Defined by this Document
Each Object defined in this document that does not allow arbitrary
additional members ends with a CDDL group socket, so that other
extensions can add parameters to it. Each Event Object type defined
in this document also ends with a CDDL group socket. The following
CDDL group sockets are defined in this document.
+===================+=========================+===============+
| Object | CDDL Socket | Defined in |
+===================+=========================+===============+
| File_Attribution | $$file-attribution-ext | Section 4.2.3 |
+-------------------+-------------------------+---------------+
| File | $$file-ext | Section 4.2.3 |
+-------------------+-------------------------+---------------+
| File_Conversation | $$file-conversation-ext | Section 4.2.3 |
+-------------------+-------------------------+---------------+
| Range | $$range-ext | Section 4.2.3 |
+-------------------+-------------------------+---------------+
| Contributor | $$contributor-ext | Section 4.2.3 |
+-------------------+-------------------------+---------------+
| Resource | $$resource-ext | Section 4.2.3 |
+-------------------+-------------------------+---------------+
| "tool-call" type | $$tool-call-event-ext | Section 4.6.1 |
| Event Object | | |
+-------------------+-------------------------+---------------+
| "tool-result" | $$tool-result-event-ext | Section 4.6.2 |
| type Event Object | | |
+-------------------+-------------------------+---------------+
| "reasoning" type | $$reasoning-event-ext | Section 4.6.3 |
| Event Object | | |
+-------------------+-------------------------+---------------+
| "system-event" | $$system-event-ext | Section 4.6.4 |
| type Event Object | | |
+-------------------+-------------------------+---------------+
Table 1: CDDL Group Extension Points Defined by this Document
Editor's Note: The Recording_Agent, Agent_Meta, Environment,
Token_Usage, and VCS_Context Objects allow arbitrary additional
members (* tstr => any), as in [VACR]. Whether these Objects get
a CDDL group socket in addition to or instead of this wildcard is
for further discussion.
4.2. vCon Object Parameters
This extension adds the following parameters to the vCon Object,
using the vcon-ext socket:
Birkholz, et al. Expires 6 May 2027 [Page 10]
Internet-Draft vCon AI Agent Extension November 2026
$$vcon-ext //= ( recording_agent: recording_agent_type )
$$vcon-ext //= ( vcs: vcs_context_type )
$$vcon-ext //= ( file_attribution: file_attribution_type )
; Recording_Agent Object
recording_agent_type = {
name: tstr,
? version: tstr,
* tstr => any
}
; File_Attribution Object
file_attribution_type = {
files: [* file_type],
* $$file-attribution-ext
}
; File Object
file_type = {
; relative path from the repository root
path: tstr,
conversations: [* file_conversation_type],
* $$file-ext
}
; File_Conversation Object
file_conversation_type = {
? url: uri_type,
; default contributor for all ranges
? contributor: contributor_type,
ranges: [* range_type],
? related: [* resource_type],
* $$file-conversation-ext
}
; Range Object
range_type = {
; 1-indexed
start_line: uint,
; 1-indexed, inclusive
end_line: uint,
? content_hash: tstr,
; default: "sha-256"
? content_hash_alg: tstr,
; overrides the conversation level contributor
? contributor: contributor_type,
* $$range-ext
}
Birkholz, et al. Expires 6 May 2027 [Page 11]
Internet-Draft vCon AI Agent Extension November 2026
; Contributor Object
contributor_type = {
type: "human" / "ai" / "mixed" / "unknown",
? model_id: tstr,
* $$contributor-ext
}
; Resource Object
resource_type = {
type: tstr,
url: uri_type,
* $$resource-ext
}
; URI (RFC 3986)
uri_type = tstr .regexp
"(([^:/?#]+):)?(//([^/?#]*))?([^?#]*)(\\?([^#]*))?(#(.*))?"
Figure 1: CDDL for the ai_agent Parameters of the vCon Object
Editor's Note: Alternative placements for recording_agent are a
Party Object that represents the recording tool and the Attachment
Object. An alternative placement for file_attribution is the
Attachment Object, as proposed in [HOWE-AGENT-SESSION].
4.2.1. recording_agent
The recording_agent parameter identifies the tool or agent that
generated this vCon, as distinct from the agent that conducted the
conversation. This distinction matters for provenance chains: the
recording tool's version affects how native data is translated into
the canonical data definition specified in this document.
* recording_agent: "Recording_Agent" (optional)
The Recording_Agent Object contains the following members:
name: The name of the recording tool or agent.
version: The version of the recording tool.
4.2.2. vcs
The vcs parameter provides version control metadata at the record
level (repository, branch, revision).
* vcs: "VCS_Context" (optional)
Birkholz, et al. Expires 6 May 2027 [Page 12]
Internet-Draft vCon AI Agent Extension November 2026
The VCS_Context Object is defined in Section 4.7.1.
4.2.3. file_attribution
NOTE: This section is specified but not yet validated against real
session data. Implementation or validation against a healthy example
corpus is pending.
File attribution captures what code was produced: which files were
modified, which line ranges were changed, and who authored them.
* file_attribution: "File_Attribution" (optional)
The File_Attribution Object is the top-level container for file
attribution data. It holds an array of files, each with their
attributed line ranges and contributor information.
files: Array of File Objects with attributed ranges.
$$file-attribution-ext: A CDDL socket that can be used to extend the
File_Attribution Object. See Section 4.1.
A File Object represents a single source file that was modified
during the conversation. It groups all conversations that
contributed changes to this file.
path: The file path relative to the repository root.
conversations: The conversations that contributed modifications to
this file, as an array of File_Conversation Objects.
$$file-ext: A CDDL socket that can be used to extend the File
Object. See Section 4.1.
A File_Conversation Object links a specific session to the line
ranges it produced in a file. This enables tracing from a line of
code back to the conversation that generated it.
url: A URL pointing to the conversation source (e.g., a web UI
permalink).
contributor: The default contributor for all ranges in this
conversation. Can be overridden per-range.
ranges: The line ranges in the file that were produced by this
conversation.
related: External resources related to this conversation (e.g.,
Birkholz, et al. Expires 6 May 2027 [Page 13]
Internet-Draft vCon AI Agent Extension November 2026
issue trackers, documentation).
$$file-conversation-ext: A CDDL socket that can be used to extend
the File_Conversation Object. See Section 4.1.
A Range Object identifies a contiguous block of lines in a file that
were produced by a specific conversation. Line numbers are 1-indexed
and inclusive. The optional content hash enables position-
independent tracking when lines move due to later edits.
start_line: The first line of the range (1-indexed).
end_line: The last line of the range (1-indexed, inclusive).
content_hash: A hash of the range content for position-independent
identification.
content_hash_alg: The hash algorithm used (default: "sha-256").
contributor: Overrides the conversation-level contributor for this
specific range.
$$range-ext: A CDDL socket that can be used to extend the Range
Object. See Section 4.1.
A Contributor Object identifies who authored a range of code. The
type member distinguishes between human-authored, AI-generated,
mixed, and unknown authorship.
type: The authorship category.
model_id: The model identifier for AI-authored ranges.
$$contributor-ext: A CDDL socket that can be used to extend the
Contributor Object. See Section 4.1.
A Resource Object represents an external reference related to a
conversation, such as an issue tracker entry, a pull request, or a
documentation page.
type: The resource type (e.g., "issue", "pr", "documentation").
url: The URL of the resource.
$$resource-ext: A CDDL socket that can be used to extend the
Resource Object. See Section 4.1.
Editor's Note: The content_hash member of the Range Object is not a
Birkholz, et al. Expires 6 May 2027 [Page 14]
Internet-Draft vCon AI Agent Extension November 2026
ContentHash as defined in [VCON-CORE]. Aligning both is for
further discussion.
4.3. Party Object Parameters
This extension adds the following parameter to the Party Object,
using the party-ext socket:
$$party-ext //= ( agent_meta: agent_meta_type )
; Agent_Meta Object
agent_meta_type = {
; e.g., "claude-opus-4-5-20251101"
model_id: tstr,
; e.g., "anthropic", "google"
model_provider: tstr,
; all models (multi-model sessions)
? models: [* tstr],
; e.g., "claude-code", "gemini-cli"
? cli_name: tstr,
? cli_version: tstr,
* tstr => any
}
Figure 2: CDDL for the ai_agent Parameters of the Party Object
Editor's Note: Alternative placements for agent_meta are the Session
Object and the generation provenance of [HOWE-PROVENANCE].
4.3.1. agent_meta
The agent_meta parameter identifies the coding agent and language
model that a Party Object represents. Agent identification is
essential for provenance tracking: knowing which model produced which
output enables auditing, capability assessment, and compliance
verification. The type parameter of a Party Object that contains
agent_meta SHOULD be "bot".
* agent_meta: "Agent_Meta" (optional)
The Agent_Meta Object contains the following members:
model_id: The primary language model identifier, using the naming
convention of the provider (e.g., "claude-opus-4-5-20251101",
"gemini-2.0-flash").
model_provider: The provider of the primary model (e.g.,
"anthropic", "google", "openai").
Birkholz, et al. Expires 6 May 2027 [Page 15]
Internet-Draft vCon AI Agent Extension November 2026
models: List of all model identifiers used during the session.
Relevant for multi-model sessions where the agent switches between
models.
cli_name: The name of the CLI tool or agent framework (e.g.,
"claude-code", "gemini-cli", "codex-cli").
cli_version: The version of the CLI tool.
4.4. Session Object Parameters
An agent session is represented by a Session Object. It captures the
full conversation between a user and an autonomous agent. The Dialog
Objects and Event Objects that are part of the Session Object
represent messages, tool invocations, reasoning steps, and system
events. Together, they preserve the complete interaction history
including all native agent metadata, enabling both replay and audit
of the conversation. The start parameter of the Session Object
provides the time the agent session began, and the duration parameter
provides the time until the agent session ended. An agent that
spawns sub-agents is represented by a Session Object that references
the Session Objects of the sub-agents using the Session_Reference
type "child", as described for AI agents in [VCON-CORE].
This extension adds the following parameters to the Session Object,
using the session-ext socket:
$$session-ext //= ( agent_session_id: tstr )
; "interactive" / "autonomous" / vendor specific
$$session-ext //= ( format: tstr )
$$session-ext //= ( environment: environment_type )
; Environment Object
environment_type = {
working_dir: tstr,
? vcs: vcs_context_type,
; sandbox mount paths
? sandboxes: [* tstr],
* tstr => any
}
Figure 3: CDDL for the ai_agent Parameters of the Session Object
Birkholz, et al. Expires 6 May 2027 [Page 16]
Internet-Draft vCon AI Agent Extension November 2026
4.4.1. agent_session_id
The agent_session_id parameter contains an opaque string uniquely
identifying the agent session in the native format of the agent
implementation. Values may be UUID v4, UUID v7, SHA-256 hashes, or
other formats. The agent_session_id parameter is distinct from the
session_id parameter of the Session Object, which contains a
SessionId Object.
* agent_session_id: "String" (optional)
4.4.2. format
The format parameter classifies the session style, such as
"interactive" (human-in-the-loop) or "autonomous" (fully automated).
Informative only; does not change the structure.
* format: "String" (optional)
4.4.3. environment
The environment parameter captures execution context for the
conversation: where the agent was running, what version control state
was active, and whether sandboxing was in effect. This context is
important for reproducibility and for understanding the scope of file
modifications.
* environment: "Environment" (optional)
The Environment Object contains the following members:
working_dir: The primary working directory path. File paths in tool
calls are typically relative to this directory.
vcs: Version control state at the time of the session (branch,
revision, etc.), as a VCS_Context Object (Section 4.7.1).
sandboxes: Paths to sandbox mount points. Some agents run in
sandboxed environments where the working directory is a temporary
mount.
Birkholz, et al. Expires 6 May 2027 [Page 17]
Internet-Draft vCon AI Agent Extension November 2026
4.5. Dialog Object Parameters
A message of an agent session represents a conversational turn:
either human input or agent response. This is the most common entry
type, carrying the primary dialog content of a session. A message is
represented by a "text" type Dialog Object. The parties parameter of
the Dialog Object references the Party Object of the contributor of
the message. Human (or upstream agent) input is contributed by the
Party Object of the user, and an agent response is contributed by the
Party Object of the agent. The message body is contained in the
Dialog Content. For assistant messages, additional metadata may be
present: the model that generated the response and token usage
statistics.
This extension adds the following parameters to the Dialog Object,
using the dialog-ext socket:
$$dialog-ext //= ( model_id: tstr )
$$dialog-ext //= ( token_usage: token_usage_type )
$$dialog-ext //= ( entry_id: agent_entry_id_type )
$$dialog-ext //= ( parent_id: agent_entry_id_type )
; Token_Usage Object
token_usage_type = {
; input tokens
? input: uint,
; output tokens
? output: uint,
; cached input tokens
? cached: uint,
; reasoning or thinking tokens
? reasoning: uint,
; total tokens
? total: uint,
; monetary cost in US dollars
? cost: number,
* tstr => any
}
Figure 4: CDDL for the ai_agent Parameters of the Dialog Object
Editor's Note: An alternative placement for model_id and token_usage
is the generation provenance of [HOWE-PROVENANCE].
4.5.1. model_id
The model_id parameter identifies the model that generated this
response. Present on assistant messages, absent on user messages.
Birkholz, et al. Expires 6 May 2027 [Page 18]
Internet-Draft vCon AI Agent Extension November 2026
* model_id: "String" (optional)
4.5.2. token_usage
The token_usage parameter captures token consumption metrics for a
model response. Token usage data is essential for cost tracking,
quota management, and understanding model behavior. All members are
optional because different agents report different subsets of token
metrics.
* token_usage: "Token_Usage" (optional)
Editor's Note: add a non-empty generic here
The Token_Usage Object contains the following members:
input: The number of input tokens consumed by this response.
output: The number of output tokens generated.
cached: The number of input tokens served from cache rather than
reprocessed.
reasoning: Tokens consumed by chain-of-thought or reasoning
computation.
total: The total token count.
cost: The monetary cost in US dollars for this response.
4.5.3. entry_id and parent_id
The entry_id parameter contains the unique identifier of this entry
within the session. The parent_id parameter references the parent
entry, enabling tree-structured conversations. Both are defined in
Section 4.7.2.
* entry_id: "String" (optional)
* parent_id: "String" (optional)
4.6. Event Object Types and Parameters
This extension defines four Event Object types representing the kinds
of entries in an agent conversation that are not messages. They are
added using the event-type-ext socket. Each type ends with a socket
of its own, so that other extensions can add parameters to it.
Birkholz, et al. Expires 6 May 2027 [Page 19]
Internet-Draft vCon AI Agent Extension November 2026
$$event-type-ext //= (event_tool_call_type,)
$$event-type-ext //= (event_tool_result_type,)
$$event-type-ext //= (event_reasoning_type,)
$$event-type-ext //= (event_system_event_type,)
; "tool-call" type Event Object
event_tool_call_type = (
type: "tool-call",
event_common_type,
; e.g., "Bash", "Edit", "Read"
tool_name: tstr,
; tool arguments
input: any,
; links call and result
? call_id: tstr,
agent_entry_reference_type,
* $$tool-call-event-ext
)
; "tool-result" type Event Object
event_tool_result_type = (
type: "tool-result",
event_common_type,
; tool output
output: any,
; links call and result
? call_id: tstr,
; "success" / "error" / "completed"
? status: tstr,
? is_error: bool,
agent_entry_reference_type,
* $$tool-result-event-ext
)
; "reasoning" type Event Object
event_reasoning_type = (
type: "reasoning",
event_common_type,
; plaintext reasoning or structured content
content: any,
; provider protected content
? encrypted: tstr,
; topic label
? subject: tstr,
agent_entry_reference_type,
* $$reasoning-event-ext
)
Birkholz, et al. Expires 6 May 2027 [Page 20]
Internet-Draft vCon AI Agent Extension November 2026
; "system-event" type Event Object
event_system_event_type = (
type: "system-event",
event_common_type,
; event classifier
event_type: tstr,
; event specific payload
? data: { * tstr => any },
agent_entry_reference_type,
* $$system-event-ext
)
Figure 5: CDDL for the ai_agent Event Object Types
For all four types, the time parameter provides when the entry
occurred, the session parameter references the Session Object of the
agent session, and the party parameter references the Party Object of
the agent. The entry_id and parent_id parameters (Section 4.7.2)
preserve the native entry identifiers and the tree structure of the
agent trace.
Editor's Note: An alternative placement for the agent trace is a
single Analysis Object that carries the trace, as proposed in
[HOWE-AGENT-SESSION]. It is also an open question whether the
"tool-call" and "reasoning" Event Object types apply to human
parties.
4.6.1. tool-call
A "tool-call" type Event Object represents a tool invocation: which
tool was called and with what arguments. Tool calls are central to
agent conversation records because tool use is the primary mechanism
by which agents interact with external environment.
tool_name: The tool name (e.g., "Bash", "Edit", "Read",
"apply_patch").
input: The arguments passed to the tool, preserved in their native
structure.
call_id: Links this call to its corresponding result.
$$tool-call-event-ext: A CDDL socket that can be used to extend the
"tool-call" type Event Object. See Section 4.1.
Birkholz, et al. Expires 6 May 2027 [Page 21]
Internet-Draft vCon AI Agent Extension November 2026
4.6.2. tool-result
A "tool-result" type Event Object represents the output returned by a
tool after execution. It is linked to its corresponding "tool-call"
type Event Object via the call_id parameter. The result carries the
tool's response data and optional status metadata indicating success
or failure.
output: The tool's response, preserved in native structure.
call_id: Links this result to its corresponding call.
status: Outcome status of the tool execution, such as "success",
"error", or "completed".
is_error: Boolean error flag, present when the tool execution
failed.
$$tool-result-event-ext: A CDDL socket that can be used to extend
the "tool-result" type Event Object. See Section 4.1.
4.6.3. reasoning
A "reasoning" type Event Object captures CoT, thinking, or internal
reasoning content from the agent. Not all agents expose reasoning
traces. When they do, the content may be plaintext, structured
blocks, or encrypted. Reasoning entries are valuable for auditing
decision-making processes and understanding why an agent took
particular actions.
content: The reasoning text or structured content. May be an empty
string when only encrypted content is available.
encrypted: Encrypted reasoning content. Used where the model
provider encrypts chain-of-thought output.
subject: A topic label for the reasoning block.
$$reasoning-event-ext: A CDDL socket that can be used to extend the
"reasoning" type Event Object. See Section 4.1.
4.6.4. system-event
A "system-event" type Event Object records system lifecycle events
that are not part of the conversation dialog but are relevant for
understanding the session context. Examples include session start
and end markers, token usage summaries, permission changes, and
configuration events.
Birkholz, et al. Expires 6 May 2027 [Page 22]
Internet-Draft vCon AI Agent Extension November 2026
event_type: Classifies the event, such as "session-start", "session-
end", "token-count", or "permission-change". The values are not
enumerated in the schema to accommodate vendor-specific event
types.
data: Event-specific payload. The structure varies by event type.
$$system-event-ext: A CDDL socket that can be used to extend the
"system-event" type Event Object. See Section 4.1.
4.6.5. Event Object Parameter Applicability by Type
The following table summarizes which Event Object parameters apply to
each Event Object type defined in this document. The symbols used in
the table are the ones used in [VCON-CORE]: M the parameter MUST be
present, O the parameter is optional, X the parameter MUST NOT be
present.
+==================+=========+===========+=========+==============+
| Parameter |tool-call|tool-result|reasoning| system-event |
+==================+=========+===========+=========+==============+
| type |M |M |M | M |
+------------------+---------+-----------+---------+--------------+
| time |M |M |M | M |
+------------------+---------+-----------+---------+--------------+
| session |M |M |M | M |
+------------------+---------+-----------+---------+--------------+
| party |M |M |M | M |
+------------------+---------+-----------+---------+--------------+
| dialog |X |X |X | X |
+------------------+---------+-----------+---------+--------------+
| button, |X |X |X | X |
| transferee, | | | | |
| transfer_target, | | | | |
| consultation, | | | | |
| target_session | | | | |
+------------------+---------+-----------+---------+--------------+
| tool_name |M |X |X | X |
+------------------+---------+-----------+---------+--------------+
| input |M |X |X | X |
+------------------+---------+-----------+---------+--------------+
| call_id |O |O |X | X |
+------------------+---------+-----------+---------+--------------+
| output |X |M |X | X |
+------------------+---------+-----------+---------+--------------+
| status |X |O |X | X |
+------------------+---------+-----------+---------+--------------+
| is_error |X |O |X | X |
Birkholz, et al. Expires 6 May 2027 [Page 23]
Internet-Draft vCon AI Agent Extension November 2026
+------------------+---------+-----------+---------+--------------+
| content |X |X |M | X |
+------------------+---------+-----------+---------+--------------+
| encrypted |X |X |O | X |
+------------------+---------+-----------+---------+--------------+
| subject |X |X |O | X |
+------------------+---------+-----------+---------+--------------+
| event_type |X |X |X | M |
+------------------+---------+-----------+---------+--------------+
| data |X |X |X | O |
+------------------+---------+-----------+---------+--------------+
| entry_id |O |O |O | O |
+------------------+---------+-----------+---------+--------------+
| parent_id |O |O |O | O |
+------------------+---------+-----------+---------+--------------+
Table 2: Event Object Parameter Applicability for the ai_agent
Event Object Types
4.7. Reusable Definitions
The CDDL definitions in this section are used by more than one of the
extension points above.
vcs_context_type = {
; "git" / "jj" / "hg" / "svn"
type: tstr,
; commit SHA or change ID
? revision: tstr,
? branch: tstr,
; repository URL
? repository: tstr,
* tstr => any
}
; Reference to an entry of the native agent trace
agent_entry_reference_type = (
? entry_id: agent_entry_id_type,
? parent_id: agent_entry_id_type
)
; Per-entry unique reference within a session
agent_entry_id_type = tstr
Figure 6: CDDL for Definitions Used by More Than One Extension Point
Birkholz, et al. Expires 6 May 2027 [Page 24]
Internet-Draft vCon AI Agent Extension November 2026
4.7.1. VCS_Context Object
The VCS_Context Object captures version control metadata for
reproducibility. Knowing the exact repository, branch, and commit at
the time of a conversation enables consumers to reconstruct the
codebase state and verify file attributions. It is used by the vcs
parameter of the vCon Object (Section 4.2.2) and by the Environment
Object of the Session Object (Section 4.4.3).
type: The version control system type (e.g., "git", "jj", "hg",
"svn").
revision: The commit SHA or change identifier at session time.
branch: The active branch name.
repository: The repository URL.
4.7.2. Agent Entry Reference
Agent implementations identify each entry of their native trace and
arrange entries in a tree. The entry_id and parent_id parameters
preserve this structure in the Dialog Objects (Section 4.5) and the
Event Objects (Section 4.6) of an agent session.
entry_id: A per-entry unique reference within a session, enabling
parent-child linking.
parent_id: References the parent entry, enabling tree-structured
conversations. Entries that the native format nests inside
another entry reference that entry in parent_id.
5. Compatibility Considerations
This extension is Compatible as defined in [VCON-CORE].
The parameters added by this extension to the vCon Object
(recording_agent, vcs, file_attribution), to the Party Object
(agent_meta), to the Session Object (agent_session_id, format,
environment), and to the Dialog Object (model_id, token_usage,
entry_id, parent_id) are new and optional. None of them changes the
meaning, structure, or permitted values of any parameter defined in
[VCON-CORE]. An implementation that does not recognize these
parameters can therefore continue to process the vCon correctly by
ignoring them.
Birkholz, et al. Expires 6 May 2027 [Page 25]
Internet-Draft vCon AI Agent Extension November 2026
The Event Object types added by this extension ("tool-call", "tool-
result", "reasoning", "system-event") are values that an
implementation that does not support this extension does not
recognize. As defined in [VCON-CORE], such an implementation ignores
these Event Objects, and they remain in the events array so that
indexes are not changed. The effect is that the agent trace is not
processed, while the messages of the agent session, which are Dialog
Objects, are processed as usual. This does not make the extension
Incompatible.
6. Incompatibility Considerations
This extension introduces no incompatibilities. All parameters and
Event Object types defined in this document are new and additive.
None of them changes the semantics, structure, or permitted values of
any parameter defined in [VCON-CORE]. Because no incompatibility is
introduced, the "ai_agent" extension name is not required to appear
in the critical parameter.
A redactor that does not support this extension cannot assess the
content of the parameters and Event Object types defined in this
document, which can carry PII (see Section 7). As described in
[VCON-CORE], such a redactor may reject a vCon that lists "ai_agent"
in the extensions parameter.
7. Privacy and Integrity Considerations
Verifiable agent conversation records reveal substantial information
about agent behavior, system state, and user interactions. The
privacy considerations of [RFC6973] apply.
7.1. Information Disclosure
User prompts captured in Dialog Objects may contain personal
identifiers, business confidential information, credentials
inadvertently included in prompts, or behavioral patterns. Agent
responses and reasoning traces may reveal inference results that
expose information about users not explicitly provided, confidential
information retrieved by tools, or system architecture details
through tool names and parameters. Implementations MUST treat user-
provided content as potentially containing personally identifiable
information.
Birkholz, et al. Expires 6 May 2027 [Page 26]
Internet-Draft vCon AI Agent Extension November 2026
Record metadata exposes operational details that may have privacy
implications. Model identifiers reveal AI capabilities, CLI versions
enable targeted attacks against known vulnerabilities, working
directories expose file system structure, and VCS context discloses
repository names and commit timing. Token usage patterns may enable
inference about conversation content even when the content itself is
protected.
7.2. Sensitive Content in Records
Reasoning entries may contain inferences that constitute special
category data, including health-related inferences from user queries,
political opinions derived from conversation context, or biometric
data processed by AI systems. The encrypted parameter of the
"reasoning" Event Object type reflects that some model providers
encrypt chain-of-thought content; when reasoning is encrypted, audit
capabilities depend on the provider's cooperation.
Tool inputs and outputs warrant particular attention. Tool call
inputs may contain credentials, API keys, or file contents. Tool
results may contain query results with personal data from databases
or external services. Implementations SHOULD provide configurable
redaction rules for common patterns and support selective entry type
recording based on deployment requirements.
7.3. Redaction and De-identification
The input parameter of "tool-call" type Event Objects, the output
parameter of "tool-result" type Event Objects, the content and
encrypted parameters of "reasoning" type Event Objects, and the data
parameter of "system-event" type Event Objects can carry party-
identifying data or PII. The same applies to the Dialog Content of
the Dialog Objects that represent messages, to the working_dir and
sandboxes members of the Environment Object, and to the path and url
members of the File_Attribution Object. A redactor is expected to
handle these parameters like Dialog Content, using the redacted
parameter of [VCON-CORE]. A redactor that removes an Event Object
entirely leaves a placeholder in the events array, so that indexes in
other Objects do not change.
7.4. External References
This extension does not add url and content_hash parameters for
externally referenced files as defined in [VCON-CORE], and therefore
introduces no new externally referenced content. The url members of
the File_Conversation Object and of the Resource Object are
references to related resources and not content of the vCon. They
are not protected by a content_hash.
Birkholz, et al. Expires 6 May 2027 [Page 27]
Internet-Draft vCon AI Agent Extension November 2026
7.5. Retention Considerations
Multiple frameworks impose retention requirements that may conflict
with data minimization principles. Organizations must balance
compliance obligations requiring extended retention against privacy
principles requiring timely deletion.
Compliant data management may require separating raw personal data
from audit trail metadata, implementing automated deletion for
personal data after compliance-minimum periods, and maintaining
cryptographic commitments enabling verification without retaining
content.
7.6. Correlation and Inference
Presentations of the same record to multiple parties can be
correlated by matching on the signature component. Session
identifiers enable linking of records across time, potentially
revealing long-term behavioral patterns or organizational structure.
Implementations SHOULD use unlinkable session identifiers where
correlation is not required.
Without accessing record content, observers may infer conversation
frequency and duration patterns, types of tools used, error rates, or
timing correlations with external events. Metadata exposure in
unprotected headers warrants careful consideration; the trace-
metadata in the COSE unprotected header (Appendix B) reveals session
identifiers, agent vendor, and timestamps even when the payload is
encrypted.
7.7. Authority Access
Regulatory frameworks may require provision of records to authorities
upon request. Access to records has to be logged, capturing who
accessed which records, when access occurred, what legal basis
justified access, and what data was disclosed. This creates a
recursive consideration: access logs may themselves contain personal
data requiring protection.
8. Security Considerations
The security considerations of [VCON-CORE] apply.
Birkholz, et al. Expires 6 May 2027 [Page 28]
Internet-Draft vCon AI Agent Extension November 2026
8.1. Record Integrity
The signed form of a vCon [VCON-CORE] and the signed-agent-record
envelope (Appendix B) provide cryptographic integrity protection for
the serialized record payload. A valid signature establishes that
the claimed signer produced the record but does not guarantee the
truthfulness of its contents. Modifications to the record structure
after signing invalidate the signature, but modifications before
signing cannot be detected. Implementations generating records
incrementally during a conversation MUST sign only after the
conversation concludes or at defined checkpoints.
8.2. Signing Key Protection
The security of signed records depends critically on the protection
of private signing keys. If an attacker obtains a signing key, they
can forge records indistinguishable from genuine ones. Protection
mechanisms range from operating system process isolation in
development environments to hardware security modules with physical
tampering resistance in high-assurance deployments. Compromised
signing keys require rejection of records signed after the compromise
date and notification to Relying Parties.
Key provisioning processes must guarantee that exclusively valid
attestation key material is established. Off-device key generation
requires confidentiality protection during transmission, creating
recursive security dependencies. On-device key generation eliminates
transmission risks but requires chain-of-custody integrity to prevent
attackers from obtaining endorsement for keys they control.
8.3. Timestamp Integrity
Timestamps establish temporal ordering of events within records.
Attackers who can manipulate timestamps can backdate records, freeze
participants in chosen time periods to evade freshness checks, or
manipulate perceived temporal relationships between entries.
Timestamps within records are attested by the signer, not
independently verified. Implementations requiring independent
timestamp verification SHOULD use external timestamping services or
transparency logs such as those defined in [RFC9943].
Birkholz, et al. Expires 6 May 2027 [Page 29]
Internet-Draft vCon AI Agent Extension November 2026
8.4. Adversarial Content
Processing verifiable agent conversation records involves parsing
content that may be produced by adversaries. This applies both to
user-supplied prompts and to model outputs that may have been
influenced by adversarial inputs. The open extensibility (* tstr =>
any) in the Objects defined in this document allows arbitrary
additional fields; implementations MUST NOT assume that unrecognized
fields are safe to process or display.
Records may contain content designed to exploit downstream systems.
Malicious prompts preserved in records may execute if records are
subsequently processed by AI systems. Tool results may contain
executable content that executes if rendered unsafely in web
interfaces. File paths in tool calls or file attribution entries may
attempt directory traversal. Implementations MUST apply appropriate
sanitization before rendering content, executing it in agent
contexts, or using paths for file system operations.
8.5. Non-Repudiation
The signed form of a vCon and the signed-agent-record envelope alone
provide authenticity and integrity, not non-repudiation. A signer
can claim key compromise or dispute the signing time without
independent evidence. Non-repudiation requires additional
infrastructure such as transparency services [RFC9943] that provide
independent timestamp proof via registration receipts, append-only
logs preventing retroactive denial, and third-party witnesses to the
signing event.
8.6. Detection and Trust Boundaries
Verifiable agent conversation records primarily enable detection of
anomalous behavior rather than prevention. Detection requires that
records accurately reflect actual agent behavior; an agent that
controls its own recording can omit or falsify entries. The
recording_agent parameter distinguishes the recording tool from the
conversing agent, enabling Relying Parties to assess trust in the
recording process.
Trust boundaries exist between the agent runtime and recording
system, between the recording system and storage, between storage and
verification, and between verification and decision-making. Attacks
at any boundary may compromise record integrity or confidentiality.
Birkholz, et al. Expires 6 May 2027 [Page 30]
Internet-Draft vCon AI Agent Extension November 2026
9. IANA Considerations
9.1. vCon JSON Registry Additions
9.1.1. vCon Extensions Names Registry
The following extension name is added to the vCon Extensions Names
Registry.
+===========+======================+============+===============+
| Extension | Extension | Change | Specification |
| Name | Description | Controller | Document(s) |
+===========+======================+============+===============+
| ai_agent | AI Agent | IESG | Section 3 RFC |
| | conversation records | | XXXX |
+-----------+----------------------+------------+---------------+
Table 3
9.1.2. vCon Object Parameter Names Registry
The following defines additional values for the vCon Object Parameter
Names Registry.
+==================+=================+============+===============+
| Parameter Name | Parameter | Change | Specification |
| | Description | Controller | Document(s) |
+==================+=================+============+===============+
| recording_agent | tool that | IESG | Section 4.2.1 |
| | generated the | | RFC XXXX |
| | record | | |
+------------------+-----------------+------------+---------------+
| vcs | record-level | IESG | Section 4.2.2 |
| | version control | | RFC XXXX |
| | context | | |
+------------------+-----------------+------------+---------------+
| file_attribution | attribution of | IESG | Section 4.2.3 |
| | modified files | | RFC XXXX |
+------------------+-----------------+------------+---------------+
Table 4
9.1.3. Party Object Parameter Names Registry
The following defines additional values for the Party Object
Parameter Names Registry.
Birkholz, et al. Expires 6 May 2027 [Page 31]
Internet-Draft vCon AI Agent Extension November 2026
+================+================+============+===============+
| Parameter Name | Parameter | Change | Specification |
| | Description | Controller | Document(s) |
+================+================+============+===============+
| agent_meta | agent and | IESG | Section 4.3.1 |
| | model metadata | | RFC XXXX |
+----------------+----------------+------------+---------------+
Table 5
9.1.4. Session Object Parameter Names Registry
The following defines additional values for the Session Object
Parameter Names Registry.
+==================+==============+============+===============+
| Parameter Name | Parameter | Change | Specification |
| | Description | Controller | Document(s) |
+==================+==============+============+===============+
| agent_session_id | native agent | IESG | Section 4.4.1 |
| | session | | RFC XXXX |
| | identifier | | |
+------------------+--------------+------------+---------------+
| format | agent | IESG | Section 4.4.2 |
| | session | | RFC XXXX |
| | style | | |
+------------------+--------------+------------+---------------+
| environment | agent | IESG | Section 4.4.3 |
| | execution | | RFC XXXX |
| | environment | | |
+------------------+--------------+------------+---------------+
Table 6
9.1.5. Dialog Object Parameter Names Registry
The following defines additional values for the Dialog Object
Parameter Names Registry.
Birkholz, et al. Expires 6 May 2027 [Page 32]
Internet-Draft vCon AI Agent Extension November 2026
+=============+======================+============+===============+
| Parameter | Parameter | Change | Specification |
| Name | Description | Controller | Document(s) |
+=============+======================+============+===============+
| model_id | model that generated | IESG | Section 4.5.1 |
| | the response | | RFC XXXX |
+-------------+----------------------+------------+---------------+
| token_usage | token consumption | IESG | Section 4.5.2 |
| | metrics | | RFC XXXX |
+-------------+----------------------+------------+---------------+
| entry_id | native agent entry | IESG | Section 4.7.2 |
| | identifier | | RFC XXXX |
+-------------+----------------------+------------+---------------+
| parent_id | native agent parent | IESG | Section 4.7.2 |
| | entry identifier | | RFC XXXX |
+-------------+----------------------+------------+---------------+
Table 7
9.1.6. Event Object Parameter Names Registry
The following defines additional values for the Event Object
Parameter Names Registry.
Birkholz, et al. Expires 6 May 2027 [Page 33]
Internet-Draft vCon AI Agent Extension November 2026
+================+=====================+============+===============+
| Parameter | Parameter | Change | Specification |
| Name | Description | Controller | Document(s) |
+================+=====================+============+===============+
| tool_name | name of the invoked | IESG | Section 4.6.1 |
| | tool | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| input | tool arguments | IESG | Section 4.6.1 |
| | | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| call_id | tool call and | IESG | Section 4.6.1 |
| | result correlation | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| output | tool output | IESG | Section 4.6.2 |
| | | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| status | tool execution | IESG | Section 4.6.2 |
| | status | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| is_error | tool execution | IESG | Section 4.6.2 |
| | error flag | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| content | reasoning content | IESG | Section 4.6.3 |
| | | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| encrypted | encrypted reasoning | IESG | Section 4.6.3 |
| | content | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| subject | reasoning topic | IESG | Section 4.6.3 |
| | label | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| event_type | system event | IESG | Section 4.6.4 |
| | classifier | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| data | system event | IESG | Section 4.6.4 |
| | payload | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| entry_id | native agent entry | IESG | Section 4.7.2 |
| | identifier | | RFC XXXX |
+----------------+---------------------+------------+---------------+
| parent_id | native agent parent | IESG | Section 4.7.2 |
| | entry identifier | | RFC XXXX |
+----------------+---------------------+------------+---------------+
Table 8
Birkholz, et al. Expires 6 May 2027 [Page 34]
Internet-Draft vCon AI Agent Extension November 2026
9.1.7. Event Type Name Registry
The following defines additional values for the Event Type Name
Registry.
+=================+=================+============+===============+
| Event Type Name | Event Type | Change | Specification |
| | Description | Controller | Document(s) |
+=================+=================+============+===============+
| tool-call | agent tool | IESG | Section 4.6.1 |
| | invocation | | RFC XXXX |
+-----------------+-----------------+------------+---------------+
| tool-result | agent tool | IESG | Section 4.6.2 |
| | result | | RFC XXXX |
+-----------------+-----------------+------------+---------------+
| reasoning | agent reasoning | IESG | Section 4.6.3 |
| | | | RFC XXXX |
+-----------------+-----------------+------------+---------------+
| system-event | agent system | IESG | Section 4.6.4 |
| | lifecycle event | | RFC XXXX |
+-----------------+-----------------+------------+---------------+
Table 9
9.1.8. New Object Parameter Names Registries
Editor's Note: [VCON-CORE] recommends requesting a parameter names
registry for each new Object. This extension defines the
Recording_Agent, File_Attribution, File, File_Conversation, Range,
Contributor, Resource, Agent_Meta, Environment, Token_Usage, and
VCS_Context Objects. The registries will be requested once the
assignment of these Objects to extension points is settled.
10. References
10.1. Normative References
[JSON] Bray, T., Ed., "The JavaScript Object Notation (JSON) Data
Interchange Format", STD 90, RFC 8259,
DOI 10.17487/RFC8259, December 2017,
<https://www.rfc-editor.org/rfc/rfc8259>.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/rfc/rfc2119>.
Birkholz, et al. Expires 6 May 2027 [Page 35]
Internet-Draft vCon AI Agent Extension November 2026
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
May 2017, <https://www.rfc-editor.org/rfc/rfc8174>.
[VCON-CORE]
Petrie, D., "The JSON format for vCon - Conversation Data
Container", Work in Progress, Internet-Draft, draft-ietf-
vcon-vcon-core-04, 7 September 2026,
<https://datatracker.ietf.org/doc/html/draft-ietf-vcon-
vcon-core-04>.
10.2. Informative References
[AICPA_TSC_2017]
American Institute of Certified Public Accountants and
Chartered Institute of Management Accountants, "Trust
Services Criteria for Security, Availability, Processing
Integrity, Confidentiality, and Privacy (With Revised
Points of Focus - 2022)", AICPA Trust Services Criteria,
2017, <https://www.aicpa-cima.com/resources/download/2017-
trust-services-criteria-with-revised-points-of-focus-
2022>.
[BSI_AI_FINANCE_2024]
Bundesamt fuer Sicherheit in der Informationstechnik,
"Test Criteria Catalogue for AI Systems in Finance",
BSI AI Finance Test Criteria, 2024,
<https://www.bsi.bund.de/SharedDocs/Downloads/EN/BSI/KI/
AI-Finance_Test-Criteria.pdf>.
[CDDL] Birkholz, H., Vigano, C., and C. Bormann, "Concise Data
Definition Language (CDDL): A Notational Convention to
Express Concise Binary Object Representation (CBOR) and
JSON Data Structures", RFC 8610, DOI 10.17487/RFC8610,
June 2019, <https://www.rfc-editor.org/rfc/rfc8610>.
[EMERGENT_MISALIGNMENT_2025]
Betley, J., Tan, D., Warncke, N., Sztyber-Betley, A., Bao,
X., Soto, M., Labenz, N., and O. Evans, "Emergent
Misalignment: Narrow finetuning can produce broadly
misaligned LLMs", arXiv 2502.17424, PMLR 267:4043-4068,
February 2025, <https://arxiv.org/abs/2502.17424>.
Birkholz, et al. Expires 6 May 2027 [Page 36]
Internet-Draft vCon AI Agent Extension November 2026
[ETSI_TS_104223_2025]
European Telecommunications Standards Institute, "Securing
Artificial Intelligence (SAI); Baseline Cyber Security
Requirements for AI Models and Systems", ETSI TS 104 223,
V1.1.1, April 2025, <https://www.etsi.org/deliver/
etsi_ts/104200_104299/104223/01.01.01_60/
ts_104223v010101p.pdf>.
[EU_AI_ACT_2024]
European Parliament and Council of the European Union,
"Regulation (EU) 2024/1689 (Artificial Intelligence Act)",
Official Journal of the European Union L 2024/1689,
CELEX: 32024R1689, July 2024,
<https://eur-lex.europa.eu/eli/reg/2024/1689/oj>.
[EU_CRA_2024]
European Parliament and Council of the European Union,
"Regulation (EU) 2024/2847 (Cyber Resilience Act)",
Official Journal of the European Union L 2024/2847,
CELEX: 32024R2847, November 2024,
<https://eur-lex.europa.eu/eli/reg/2024/2847/oj>.
[EU_NIS2_2022]
European Parliament and Council of the European Union,
"Directive (EU) 2022/2555 (NIS 2 Directive)", Official
Journal of the European Union L 333, CELEX: 32022L2555,
December 2022.
[FFIEC_IT_HANDBOOK_2024]
Federal Financial Institutions Examination Council,
"Information Technology Examination Handbook", FFIEC IT
Examination Handbook, August 2024,
<https://ithandbook.ffiec.gov/>.
[HOWE-AGENT-SESSION]
McCarthy-Howe, T., "vCon Agent Session", Work in Progress,
Internet-Draft, draft-howe-vcon-agent-session-00, 20 May
2026, <https://datatracker.ietf.org/doc/html/draft-howe-
vcon-agent-session-00>.
[HOWE-PROVENANCE]
McCarthy-Howe, T., "vCon Generation Provenance", Work in
Progress, Internet-Draft, draft-howe-vcon-provenance-00, 4
June 2026, <https://datatracker.ietf.org/doc/html/draft-
howe-vcon-provenance-00>.
Birkholz, et al. Expires 6 May 2027 [Page 37]
Internet-Draft vCon AI Agent Extension November 2026
[ISO_IEC_42001_2023]
International Organization for Standardization and
International Electrotechnical Commission, "Information
technology - Artificial intelligence - Management system",
ISO/IEC 42001:2023, December 2023,
<https://www.iso.org/standard/42001>.
[ISO_IEC_DIS_24970]
International Organization for Standardization and
International Electrotechnical Commission, "Artificial
intelligence - AI system logging", ISO/IEC DIS 24970,
2025, <https://www.iso.org/standard/88723.html>.
[NIST_AI_100_2_E2025]
National Institute of Standards and Technology,
"Adversarial Machine Learning: A Taxonomy and Terminology
of Attacks and Mitigations", NIST AI 100-2 E2025, March
2025, <https://csrc.nist.gov/pubs/ai/100/2/e2025/final>.
[NIST_SP_800_53_R5]
National Institute of Standards and Technology, "Security
and Privacy Controls for Information Systems and
Organizations", NIST Special Publication 800-53, Revision
5, September 2020,
<https://csrc.nist.gov/pubs/sp/800/53/r5/upd1/final>.
[OMB_M_21_31]
Office of Management and Budget, "Improving the Federal
Government's Investigative and Remediation Capabilities
Related to Cybersecurity Incidents", OMB
Memorandum M-21-31, August 2021,
<https://www.whitehouse.gov/wp-content/uploads/2021/08/
M-21-31-Improving-the-Federal-Governments-Investigative-
and-Remediation-Capabilities-Related-to-Cybersecurity-
Incidents.pdf>.
[PCI_DSS_V4_0_1]
PCI Security Standards Council, "Payment Card Industry
Data Security Standard", PCI DSS Version 4.0.1, June 2024,
<https://www.pcisecuritystandards.org/document_library/>.
[RFC6973] Cooper, A., Tschofenig, H., Aboba, B., Peterson, J.,
Morris, J., Hansen, M., and R. Smith, "Privacy
Considerations for Internet Protocols", RFC 6973,
DOI 10.17487/RFC6973, July 2013,
<https://www.rfc-editor.org/rfc/rfc6973>.
Birkholz, et al. Expires 6 May 2027 [Page 38]
Internet-Draft vCon AI Agent Extension November 2026
[RFC9334] Birkholz, H., Thaler, D., Richardson, M., Smith, N., and
W. Pan, "Remote ATtestation procedureS (RATS)
Architecture", RFC 9334, DOI 10.17487/RFC9334, January
2023, <https://www.rfc-editor.org/rfc/rfc9334>.
[RFC9360] Schaad, J., "CBOR Object Signing and Encryption (COSE):
Header Parameters for Carrying and Referencing X.509
Certificates", RFC 9360, DOI 10.17487/RFC9360, February
2023, <https://www.rfc-editor.org/rfc/rfc9360>.
[RFC9393] Birkholz, H., Fitzgerald-McKay, J., Schmidt, C., and D.
Waltermire, "Concise Software Identification Tags",
RFC 9393, DOI 10.17487/RFC9393, June 2023,
<https://www.rfc-editor.org/rfc/rfc9393>.
[RFC9597] Looker, T. and M.B. Jones, "CBOR Web Token (CWT) Claims in
COSE Headers", RFC 9597, DOI 10.17487/RFC9597, June 2024,
<https://www.rfc-editor.org/rfc/rfc9597>.
[RFC9682] Bormann, C., "Updates to the Concise Data Definition
Language (CDDL) Grammar", RFC 9682, DOI 10.17487/RFC9682,
November 2024, <https://www.rfc-editor.org/rfc/rfc9682>.
[RFC9942] Steele, O., Birkholz, H., Delignat-Lavaud, A., and C.
Fournet, "CBOR Object Signing and Encryption (COSE)
Receipts", RFC 9942, DOI 10.17487/RFC9942, June 2026,
<https://www.rfc-editor.org/rfc/rfc9942>.
[RFC9943] Birkholz, H., Delignat-Lavaud, A., Fournet, C., Deshpande,
Y., and S. Lasker, "An Architecture for Trustworthy and
Transparent Digital Supply Chains", RFC 9943,
DOI 10.17487/RFC9943, June 2026,
<https://www.rfc-editor.org/rfc/rfc9943>.
[ROSENBERG-RESTRUCTURE]
Rosenberg, J., "Virtualized Conversations (VCON)
Restructure to Facilitate AI Agent Use Cases", Work in
Progress, Internet-Draft, draft-rosenberg-vcon-
restructure-00, 11 September 2026,
<https://datatracker.ietf.org/doc/html/draft-rosenberg-
vcon-restructure-00>.
[STD94] Bormann, C. and P. Hoffman, "Concise Binary Object
Representation (CBOR)", STD 94, RFC 8949,
DOI 10.17487/RFC8949, December 2020,
<https://www.rfc-editor.org/rfc/rfc8949>.
Birkholz, et al. Expires 6 May 2027 [Page 39]
Internet-Draft vCon AI Agent Extension November 2026
[STD96] Schaad, J., "CBOR Object Signing and Encryption (COSE):
Structures and Process", STD 96, RFC 9052,
DOI 10.17487/RFC9052, August 2022,
<https://www.rfc-editor.org/rfc/rfc9052>.
[VACR] Birkholz, H., Heldt, T., and O. Steele, "Verifiable Agent
Conversation Records", Work in Progress, Internet-Draft,
draft-birkholz-verifiable-agent-conversations-01, 31
August 2026, <https://datatracker.ietf.org/doc/html/draft-
birkholz-verifiable-agent-conversations-01>.
Appendix A. Compliance Requirements for AI Agent Conversation Records
This section identifies the intersection of logging, traceability,
and record-keeping requirements across major compliance frameworks
applicable to AI systems. The verifiable agent conversation format
defined in this document addresses these requirements by providing a
standardized, cryptographically verifiable record of AI agent
interactions.
A.1. Applicable Requirement Frameworks
The following frameworks were analyzed for their requirements on AI
agent traceability and session logging:
Birkholz, et al. Expires 6 May 2027 [Page 40]
Internet-Draft vCon AI Agent Extension November 2026
+==========================+=============+=============+==========+
| Framework |Jurisdiction |Sector |Status |
+==========================+=============+=============+==========+
| EU AI Act |EU |Cross-sector |In force |
| [EU_AI_ACT_2024] | | |Aug 2024 |
+--------------------------+-------------+-------------+----------+
| Cyber Resilience Act |EU |Products with|In force |
| [EU_CRA_2024] | |digital |Dec 2024 |
| | |elements | |
+--------------------------+-------------+-------------+----------+
| NIS2 Directive |EU |Essential/ |Transposed|
| [EU_NIS2_2022] | |important |Oct 2024 |
| | |entities | |
+--------------------------+-------------+-------------+----------+
| ETSI TS 104 223 |EU/ |AI systems |Published |
| [ETSI_TS_104223_2025] |International| |Apr 2025 |
+--------------------------+-------------+-------------+----------+
| SOC 2 Trust Services |US/ |Service |Active |
| Criteria |International|organizations| |
| [AICPA_TSC_2017] | | | |
+--------------------------+-------------+-------------+----------+
| FedRAMP |US |Federal cloud|Active |
| [NIST_SP_800_53_R5] | |services | |
| [OMB_M_21_31] | | | |
+--------------------------+-------------+-------------+----------+
| PCI DSS [PCI_DSS_V4_0_1] |International|Payment card |Mandatory |
| | |industry |Mar 2025 |
+--------------------------+-------------+-------------+----------+
| ISO/IEC 42001 |International|AI management|Published |
| [ISO_IEC_42001_2023] | |systems |2023 |
+--------------------------+-------------+-------------+----------+
| FFIEC IT Handbook |US |Financial |Updated |
| [FFIEC_IT_HANDBOOK_2024] | |institutions |2024 |
+--------------------------+-------------+-------------+----------+
| BSI AI Finance Test |Germany |Financial |Published |
| Criteria | |sector AI |2024 |
| [BSI_AI_FINANCE_2024] | | | |
+--------------------------+-------------+-------------+----------+
| NIST AI 100-2 |US |Cross-sector |Published |
| [NIST_AI_100_2_E2025] | | |2025 |
+--------------------------+-------------+-------------+----------+
Table 10
Birkholz, et al. Expires 6 May 2027 [Page 41]
Internet-Draft vCon AI Agent Extension November 2026
A.2. Common Requirements Intersection
The analysis of the frameworks listed above results in eleven
categories of requirements that appear across ALL or MOST frameworks.
These categories frame and represent a minimum baseline that
verifiable agent conversation records MUST support.
A.2.1. REQ-1: Automatic Event Logging
All frameworks require automatic, system-generated logging of events
without reliance on manual recording. Explicit requirements are
listed as follows:
+===================+===============================================+
| Framework | Requirement |
+===================+===============================================+
| EU AI Act Art. | "High-risk AI systems shall technically allow |
| 12 | for the automatic recording of events (logs)" |
+-------------------+-----------------------------------------------+
| ETSI TS 104223 | "System Operators shall log system and user |
| 5.4.2-1 | actions" |
+-------------------+-----------------------------------------------+
| SOC 2 CC7.2 | "Complete and chronological record of all |
| | user actions and system responses" |
+-------------------+-----------------------------------------------+
| FedRAMP AU-12 | "Audit Record Generation" control requirement |
+-------------------+-----------------------------------------------+
| PCI DSS 4.0 | "Audit logs implemented to support detection |
| Req 10 | of anomalies" |
+-------------------+-----------------------------------------------+
| ISO 42001 | "AI system recording of event logs" |
| A.6.2.8 | |
+-------------------+-----------------------------------------------+
Table 11
Mapping to this specification: The Dialog Objects and Event Objects
of a Session Object capture all events automatically. Each of them
represents a discrete, system-recorded event with structured
metadata.
A.2.2. REQ-2: Timestamp Requirements
Most frameworks require precise temporal information for each logged
event. Explicit requirements are listed as follows:
Birkholz, et al. Expires 6 May 2027 [Page 42]
Internet-Draft vCon AI Agent Extension November 2026
+======================+====================================+
| Framework | Requirement |
+======================+====================================+
| EU AI Act Art. 12(2) | "Precise timestamps for each usage |
| | session" (biometric systems) |
+----------------------+------------------------------------+
| PCI DSS 4.0 Req 10.6 | "Time-synchronization mechanisms |
| | support consistent time settings" |
+----------------------+------------------------------------+
| SOC 2 | "When the activity was performed |
| | via timestamp" |
+----------------------+------------------------------------+
| NIS2 | "Precise logging of when an |
| | incident was first detected" |
+----------------------+------------------------------------+
Table 12
Mapping to this specification: The start parameter of each Dialog
Object and the time parameter of each Event Object use the Date type
of [VCON-CORE], which is an RFC 3339 string, ensuring
interoperability across implementations.
A.2.3. REQ-3: Actor Identification
Most frameworks require some attribution of actions to identifiable
actors (human or system). Explicit requirements are listed as
follows:
+===============+==========================================+
| Framework | Requirement |
+===============+==========================================+
| EU AI Act | "Identification of the natural persons |
| Art. 12(3)(d) | involved in the verification of results" |
+---------------+------------------------------------------+
| SOC 2 | "The process or user who initiated the |
| | activity (Who)" |
+---------------+------------------------------------------+
| PCI DSS 4.0 | Attribution to "Who" performed each |
| | action |
+---------------+------------------------------------------+
| FedRAMP AC-2 | Account management and identification |
+---------------+------------------------------------------+
Table 13
Birkholz, et al. Expires 6 May 2027 [Page 43]
Internet-Draft vCon AI Agent Extension November 2026
Mapping to this specification: The contributor type captures actor
attribution with type (human/ai/mixed/unknown) and optional model_id.
The agent_meta parameter of the Party Object identifies the AI
system.
A.2.4. REQ-4: Action/Event Type Recording
All frameworks require recording of what action or event occurred.
Explicit requirements are listed as follows:
+===================+==========================================+
| Framework | Requirement |
+===================+==========================================+
| EU AI Act Art. 12 | "Events relevant for identifying |
| | situations that may result in...risk" |
+-------------------+------------------------------------------+
| ETSI TS 104223 | "Audit log of changes to system prompts |
| 5.2.4-3 | or other model configuration" |
+-------------------+------------------------------------------+
| SOC 2 | "The action they performed such as file |
| | transferred, created, or deleted (What)" |
+-------------------+------------------------------------------+
| PCI DSS 4.0 | "What" component of audit trail |
+-------------------+------------------------------------------+
Table 14
Mapping to this specification: The type parameter of Dialog Objects
and Event Objects discriminates event types: text (messages), tool-
call, tool-result, reasoning, system-event.
A.2.5. REQ-5: Input/Output Recording (AI-Specific)
All AI-specific frameworks require recording of inputs (prompts) and
outputs (responses). Explicit requirements are listed as follows:
Birkholz, et al. Expires 6 May 2027 [Page 44]
Internet-Draft vCon AI Agent Extension November 2026
+================+==========================================+
| Framework | Requirement |
+================+==========================================+
| EU AI Act Art. | "The input data for which the search has |
| 12(3)(c) | led to a match" |
+----------------+------------------------------------------+
| PCI DSS AI | "Logging should be sufficient to audit |
| Guidance | the prompt inputs and reasoning process" |
+----------------+------------------------------------------+
| ETSI TS 104223 | "Operation, and lifecycle management of |
| 5.1.2-3 | models, datasets and prompts" |
+----------------+------------------------------------------+
| FFIEC VII.D | "Lack of explainability...unclear how |
| | inputs are translated into outputs" |
+----------------+------------------------------------------+
Table 15
Mapping to this specification:
* text Dialog Object contributed by the user: User/system input
(prompt)
* text Dialog Object contributed by the agent: Model response
* tool-call Event Object input: Tool invocation parameters
* tool-result Event Object output: Tool execution results
* reasoning Event Object content: Chain-of-thought (where available)
A.2.6. REQ-6: Retention Period Requirements
Most frameworks specify minimum retention periods for audit logs.
Explicit requirements are listed as follows:
Birkholz, et al. Expires 6 May 2027 [Page 45]
Internet-Draft vCon AI Agent Extension November 2026
+===================+============================================+
| Framework | Minimum Retention |
+===================+============================================+
| EU AI Act Art. 19 | 6 months (longer for financial services) |
+-------------------+--------------------------------------------+
| FedRAMP (M-21-31) | 12 months active + 18 months cold storage |
+-------------------+--------------------------------------------+
| PCI DSS 4.0 | 12 months total, 3 months immediate access |
+-------------------+--------------------------------------------+
| NIS2 | Per member state law |
+-------------------+--------------------------------------------+
Table 16
Recommendation: Implementations SHOULD retain verifiable agent
conversation records for at least 12 months to satisfy the most
common requirement threshold.
A.2.7. REQ-7: Tamper-Evidence and Integrity Protection
All frameworks require some protection against unauthorized
modification of logs. Explicit requirements are listed as follows:
+======================+===================================+
| Framework | Requirement |
+======================+===================================+
| PCI DSS 4.0 Req 10.5 | "Tamper-proof audit trails...logs |
| | cannot be altered retroactively" |
+----------------------+-----------------------------------+
| FedRAMP | "Effective chain of evidence to |
| | ensure integrity" |
+----------------------+-----------------------------------+
| SOC 2 | Log integrity as security control |
+----------------------+-----------------------------------+
| CRA | "Tamper-proof SBOMs and |
| | vulnerability disclosures" |
+----------------------+-----------------------------------+
Table 17
Mapping to this specification: The signed form of a vCon and the
signed-agent-record type (COSE_Sign1 envelope, Appendix B) provide
cryptographic integrity protection. The content-hash field in trace-
metadata enables verification of payload integrity.
Birkholz, et al. Expires 6 May 2027 [Page 46]
Internet-Draft vCon AI Agent Extension November 2026
A.2.8. REQ-8: Incident Response Support
All frameworks require some logs to support incident investigation
and response. Explicit requirements are listed as follows:
+================+==================================================+
| Framework | Requirement |
+================+==================================================+
| NIS2 Art. 23 | 24-hour initial notification, |
| | 72-hour assessment |
+----------------+--------------------------------------------------+
| CRA | 24-hour vulnerability notification |
| | to ENISA |
+----------------+--------------------------------------------------+
| ETSI TS 104223 | Logs for "incident investigations, |
| 5.4.2-1 | and vulnerability remediation" |
+----------------+--------------------------------------------------+
| FedRAMP | Incident reporting and continuous |
| | monitoring |
+----------------+--------------------------------------------------+
Table 18
Mapping to this specification: The structured format enables rapid
extraction of relevant entries by timestamp range, event type, or
tool invocation for incident reconstruction.
A.2.9. REQ-9: Anomaly and Risk Detection Support
All frameworks require some logs to enable detection of anomalous or
risky behavior. Explicit requirements are listed as follows:
+=========================+====================================+
| Framework | Requirement |
+=========================+====================================+
| EU AI Act Art. 12(2)(a) | "Identifying situations that may |
| | result in...risk" |
+-------------------------+------------------------------------+
| ETSI TS 104223 5.4.2-2 | "Detect anomalies, security |
| | breaches, or unexpected behaviour" |
+-------------------------+------------------------------------+
| FedRAMP SI-4 | "Anomaly detection" |
+-------------------------+------------------------------------+
| SOC 2 | "Anomaly detection" for security |
| | monitoring |
+-------------------------+------------------------------------+
Table 19
Birkholz, et al. Expires 6 May 2027 [Page 47]
Internet-Draft vCon AI Agent Extension November 2026
Mapping to this specification: The standardized entry types and
structured tool-call/tool-result pairs enable automated analysis for
detecting:
* Unusual tool invocation patterns
* Failed operations (via status and is_error fields)
* Unexpected reasoning patterns
* Token usage anomalies
A.2.10. REQ-10: Human Oversight Enablement
All AI-specific frameworks require logs to support human review and
oversight. Explicit requirements are listed as follows:
+======================+============================================+
| Framework | Requirement |
+======================+============================================+
| EU AI Act Art. | "Monitoring the operation of high-risk AI |
| 26(5) | systems" |
+----------------------+--------------------------------------------+
| ETSI TS 104223 | "Capabilities to enable human oversight" |
| 5.1.4-1 | |
+----------------------+--------------------------------------------+
| ISO 42001 | Human responsibility and accountability |
+----------------------+--------------------------------------------+
| FFIEC VII.D | "Dynamic updating...challenges to |
| | monitoring and independently reviewing AI" |
+----------------------+--------------------------------------------+
Table 20
Mapping to this specification: The reasoning Event Object type
captures chain-of-thought content (where available), enabling human
reviewers to understand AI decision-making processes. The entry_id
and parent_id fields and the hierarchy of Session Objects preserve
conversation structure.
A.2.11. REQ-11: Traceability and Reproducibility
All frameworks require some ability to trace system behavior and
reconstruct events. Explicit requirements are listed as follows:
Birkholz, et al. Expires 6 May 2027 [Page 48]
Internet-Draft vCon AI Agent Extension November 2026
+===========+====================================================+
| Framework | Requirement |
+===========+====================================================+
| EU AI Act | "Level of traceability of the |
| Art. 12 | functioning...appropriate to the intended purpose" |
+-----------+----------------------------------------------------+
| ISO 42001 | "Traceability" as key factor including "data |
| | provenance, model traceability" |
+-----------+----------------------------------------------------+
| CRA | "Traceability in the software supply chain" |
+-----------+----------------------------------------------------+
| ETSI TS | "Track, authenticate, manage version control" |
| 104223 | |
| 5.2.1-2 | |
+-----------+----------------------------------------------------+
Table 21
Mapping to this specification:
* Session Object and agent_session_id: Links entries to sessions
* entry_id and parent_id: Enables conversation tree reconstruction
* vcs_context: Git commit/branch for code state
* agent_meta: Model version and CLI version
* file_attribution: Code provenance tracking
A.3. Framework-Specific Requirements
A.3.1. EU AI Act (High-Risk Systems)
For AI systems classified as high-risk under Annex III of the EU AI
Act [EU_AI_ACT_2024], additional requirements apply:
1. Biometric identification systems (Annex III, 1(a)) require
logging of:
* Precise timestamps for start/end of each usage session
* Reference database used during input data validation
* Input data leading to matches
* Natural persons involved in result verification
Birkholz, et al. Expires 6 May 2027 [Page 49]
Internet-Draft vCon AI Agent Extension November 2026
2. Log retention: Minimum 6 months; financial services may require
longer per sector-specific regulation.
3. Authority Access: Art. 19 of [EU_AI_ACT_2024] requires provision
of logs to competent authorities upon reasoned request.
A.3.2. ETSI TS 104 223 Session Logging Requirements
ETSI TS 104 223 [ETSI_TS_104223_2025] provides the most detailed AI-
specific logging requirements:
+===========+=================================+===================+
| Provision | Requirement | This Spec Mapping |
+===========+=================================+===================+
| 5.1.2-3 | Audit trail for "operation, and | Session Object, |
| | lifecycle management of models, | agent_meta |
| | datasets and prompts" | |
+-----------+---------------------------------+-------------------+
| 5.2.4-1 | "Document and maintain a clear | recording_agent, |
| | audit trail of their system | open maps |
| | design" | |
+-----------+---------------------------------+-------------------+
| 5.2.4-3 | "Audit log of changes to system | system-event |
| | prompts or other model | Event Object with |
| | configuration" | prompt changes |
+-----------+---------------------------------+-------------------+
| 5.4.2-1 | "Log system and user actions to | Dialog Objects |
| | support security compliance, | and Event Objects |
| | incident investigations" | |
+-----------+---------------------------------+-------------------+
| 5.4.2-2 | "Analyse their logs to | Structured format |
| | ensure...desired outputs and to | enables analysis |
| | detect anomalies" | |
+-----------+---------------------------------+-------------------+
| 5.4.2-3 | "Monitor internal states of | reasoning Event |
| | their AI systems" | Object, |
| | | token_usage |
+-----------+---------------------------------+-------------------+
Table 22
A.3.3. PCI DSS AI-Specific Guidance
The PCI Security Standards Council has published guidance on AI in
payment environments [PCI_DSS_V4_0_1]: "Where possible, logging
should be sufficient to audit the prompt inputs and reasoning process
used by the AI system that led to the output provided."
Birkholz, et al. Expires 6 May 2027 [Page 50]
Internet-Draft vCon AI Agent Extension November 2026
This specification directly addresses this requirement through:
* text Dialog Object contributed by the user: Captures prompt inputs
* reasoning Event Object: Captures chain-of-thought (where
available)
* text Dialog Object contributed by the agent: Captures model
outputs
* tool-call / tool-result Event Objects: Captures agentic actions
A.3.4. Financial Sector Requirements (FFIEC, BSI)
Financial institutions face additional scrutiny for AI systems per
[FFIEC_IT_HANDBOOK_2024] and [BSI_AI_FINANCE_2024]:
+================+===========================+===================+
| Requirement | FFIEC | BSI AI Finance |
| Area | | |
+================+===========================+===================+
| Explainability | "Lack of transparency or | Test criteria for |
| | explainability" risk | explainability |
+----------------+---------------------------+-------------------+
| Dynamic | "Challenges to monitoring | Continuous |
| updating | and independently | validation |
| | reviewing AI" | |
+----------------+---------------------------+-------------------+
| Audit trail | Log management (VI.B.7) | Complete audit |
| | | trail |
+----------------+---------------------------+-------------------+
Table 23
A.4. Compliance Mapping Table
The following table maps this specification's data elements to
compliance requirements:
Birkholz, et al. Expires 6 May 2027 [Page 51]
Internet-Draft vCon AI Agent Extension November 2026
+================+========+=========+=====+=======+=====+=======+====+
|Data Element |EU AI |ETSI |SOC 2|FedRAMP|PCI |ISO |NIS2|
| |Act |104223 | | |DSS |42001 | |
+================+========+=========+=====+=======+=====+=======+====+
|start / time |Art. |5.4.2-1 |CC7.2|AU-8 |10.6 |A.6.2.8|Art.|
| |12(2) | | | | | |23 |
+----------------+--------+---------+-----+-------+-----+-------+----+
|agent_session_id|Art. 12 |5.2.4-1 |CC7.2|AU-3 |10.2 |A.6.2.8|- |
+----------------+--------+---------+-----+-------+-----+-------+----+
|type |Art. |5.4.2-1 |CC7.2|AU-3 |10.2 |A.6.2.8|- |
| |12(2) | | | | | | |
+----------------+--------+---------+-----+-------+-----+-------+----+
|contributor |Art. |5.1.4 |CC6.1|AC-2 |10.2 |A.6.2.8|- |
| |12(3)(d)| | | | | | |
+----------------+--------+---------+-----+-------+-----+-------+----+
|Dialog Content |Art. |5.1.2-3 |- |- |AI |- |- |
| |12(3)(c)| | | |Guide| | |
+----------------+--------+---------+-----+-------+-----+-------+----+
|reasoning |Art. 12 |5.4.2-3 |- |- |AI |A.7.1 |- |
| | | | | |Guide| | |
+----------------+--------+---------+-----+-------+-----+-------+----+
|tool-call / |Art. 12 |5.4.2-1 |CC7.2|AU-12 |10.2 |A.6.2.8|- |
|tool-result | | | | | | | |
+----------------+--------+---------+-----+-------+-----+-------+----+
|signed-agent- |Art. 19 |5.2.4-1.2|CC6.1|AU-9 |10.5 |- |- |
|record | | | | | | | |
+----------------+--------+---------+-----+-------+-----+-------+----+
|vcs_context |- |5.2.1-2 |- |CM-3 |- |A.6.2.8|- |
+----------------+--------+---------+-----+-------+-----+-------+----+
|token_usage |- |5.4.2-4 |- |- |- |- |- |
+----------------+--------+---------+-----+-------+-----+-------+----+
Table 24
A.5. Security Requirements addressing Regulatory Compliance
A.5.1. Log Integrity
Per PCI DSS [PCI_DSS_V4_0_1] Req 10.5 and FedRAMP [NIST_SP_800_53_R5]
AU-9, logs MUST be protected against modification. Implementations
SHOULD:
1. Use the signed form of the vCon or the signed-agent-record
envelope (Appendix B) for cryptographic integrity
2. Store the content-hash for offline verification
3. Implement write-once storage for log archives
Birkholz, et al. Expires 6 May 2027 [Page 52]
Internet-Draft vCon AI Agent Extension November 2026
A.5.2. Access Control
Per FedRAMP [NIST_SP_800_53_R5] AC-3 and ETSI [ETSI_TS_104223_2025]
5.2.2-1, access to logs MUST be controlled:
1. Logs containing sensitive prompts or outputs require access
control
2. Reasoning content may contain confidential information
3. Authority access (EU AI Act Art. 19) requires audit of log access
itself
A.5.3. Data Protection
Logs may contain personal data subject to GDPR/privacy regulations:
1. Dialog Content of text Dialog Objects may contain PII
2. output of tool-result Event Objects may contain query results
with PII
3. Retention periods must balance compliance requirements with data
minimization
A.6. Related Standards and Emerging Work
The ISO/IEC JTC 1/SC 42 committee is developing [ISO_IEC_DIS_24970],
a draft international standard specifically addressing AI system
logging requirements. This work aligns with and complements the
verifiable agent conversation format defined in this document.
Research on emergent misalignment [EMERGENT_MISALIGNMENT_2025]
demonstrates that narrow fine-tuning can produce broadly misaligned
LLMs, underscoring the importance of comprehensive conversation
logging for detecting behavioral anomalies that may not be apparent
from individual interactions.
Appendix B. Application of SCITT and COSE for Transparency and
Authenticity
The signed form of a vCon [VCON-CORE] uses JWS. This appendix
describes a signing envelope that provides cryptographic integrity
protection for vCons that carry agent conversation records using
COSE_Sign1 ([STD96]), which is encoded in CBOR [STD94]. The vCon is
the payload of the envelope, and the content type header parameter
carries the media type "application/vcon" defined in [VCON-CORE].
This makes a signed agent record a Signed Statement that can be
Birkholz, et al. Expires 6 May 2027 [Page 53]
Internet-Draft vCon AI Agent Extension November 2026
registered with a SCITT Transparency Service [RFC9943]. No extension
point of the vCon schema is used by the envelope. The reader is
assumed to be familiar with the vocabulary and concepts defined in
[RFC9334] and [RFC9943].
Editor's Note: The use of COSE and SCITT for vCons is expected to
progress together with the SCITT related vCon work.
The collated CDDL definition for the envelope is specified as
follows:
=============== NOTE: '\' line wrapping per RFC 8792 ================
signed-agent-record = #6.18([ ; COSE_Sign1 tag
protected: bstr .cbor protected-header ; {alg, content-type, \
scitt-stuff}
unprotected: unprotected-header ; Trace metadata
payload: bstr / null ; Detached if null
signature: bstr
])
protected-header = {
&(CWT_Claims: 15) => CWT_Claims
? &(alg: 1) => int
? &(content_type: 3) => tstr / uint
? &(kid: 4) => bstr
? &(x5t: 34) => COSE_CertHash
? &(x5chain: 33) => COSE_X509
* label => any
}
CWT_Claims = {
&(iss: 1) => tstr
&(sub: 2) => tstr
* label => any
}
unprotected-header = {
? &(trace-metadata-key: 100) => trace-metadata ; 100 is placeholder
? &(x5chain: 33) => COSE_X509
? &(receipts: 394) => [ + Receipt ]
* label => any
}
trace-metadata = {
session-id: session-id
agent-vendor: tstr
trace-format: trace-format-id
Birkholz, et al. Expires 6 May 2027 [Page 54]
Internet-Draft vCon AI Agent Extension November 2026
timestamp-start: abstract-timestamp
? timestamp-end: abstract-timestamp
? content-hash: tstr ; SHA-256 hex digest of \
payload
? content-hash-alg: tstr
}
; Known values: "ietf-vac-v3.0" (canonical), "claude-jsonl", "gemini\
-json",
; "codex-jsonl", "opencode-json", "cursor-jsonl". Extensible via \
tstr.
trace-format-id = tstr
COSE_X509 = bstr / [ 2*certs: bstr ]
COSE_CertHash = [ hashAlg: (int / tstr), hashValue: bstr ]
label = int / tstr
; COSE Receipt CDDL for use in SCITT compliant COSE Envelope
Receipt = #6.18(COSE_Sign1)
cose-label = int / tstr
cose-value = any
Protected_Header = {
* cose-label => cose-value
}
Unprotected_Header = {
&(receipts: 394) => [+ bstr .cbor Receipt]
* cose-label => cose-value
}
COSE_Sign1 = [
protected : bstr .cbor Protected_Header,
unprotected : Unprotected_Header,
payload : bstr / null,
signature : bstr
]
; SCITT-interoperable COSE Envelope
; including from draft-ietf-cose-merkle-tree-proofs and
; draft-ietf-scitt-architecture for validation
; RFC 3339 string OR epoch milliseconds (for interop).
abstract-timestamp = tstr .regexp date-time-regexp / uint
Birkholz, et al. Expires 6 May 2027 [Page 55]
Internet-Draft vCon AI Agent Extension November 2026
; Opaque string: UUID, SHA-256 hash in base64url, etc.
session-id = tstr / bstr
; RFC 3339 date-time pattern
date-time-regexp = "([0-9]{4})-(0[1-9]|1[0-2])-(0[1-9]|[12][0-9]|3[\
01])T([01][0-9]|2[0-3]):([0-5][0-9]):(60|[0-5][0-9])([.][0-9]+)?(Z|[\
+-]([01][0-9]|2[0-3]):[0-5][0-9])"
Figure 7: CDDL for the Signed Agent Record Envelope
B.1. The signed-agent-record Structure
A signed-agent-record is a COSE_Sign1 envelope (CBOR Tag 18) that
wraps a vCon with a cryptographic signature. Signing provides data
provenance and tamper evidence, satisfying some requirements of RATS
Evidence generation ([RFC9334]) and SCITT auditability ([RFC9943])
requirements. The payload may be included or detached (null); in
detached mode, the vCon is supplied separately during verification.
The following describes each element of this structure.
protected: The serialized protected header containing the algorithm
identifier, content type, and CWT claims [RFC9597].
unprotected: The unprotected header carrying trace metadata at label
100 and, after registration, the receipts [RFC9942] of a SCITT
Transparency Service.
payload: The serialized vCon bytes, or null for detached payloads.
signature: The cryptographic signature over the protected header and
payload.
The x5chain and x5t header parameters are defined in [RFC9360].
B.2. The trace-metadata Map
The trace-metadata type carries summary information about the signed
record in the COSE_Sign1 unprotected header. This enables consumers
to inspect key properties of a signed record without deserializing
the full payload.
The following describes each member of this map.
session-id: The session identifier from the signed record.
agent-vendor: The agent provider name (e.g., "anthropic", "google").
Birkholz, et al. Expires 6 May 2027 [Page 56]
Internet-Draft vCon AI Agent Extension November 2026
trace-format: Identifies the format of the signed payload (e.g.,
"ietf-vac-v3.0" for canonical records).
timestamp-start: When the session began.
timestamp-end: When the session ended.
content-hash: SHA-256 hex digest of the payload bytes, enabling
integrity checking independent of the COSE signature.
content-hash-alg: The hash algorithm used (default: "sha-256").
Editor's Note: The label 100 for trace-metadata is a placeholder. A
COSE Header Parameter registration is for further discussion.
Appendix C. Mapping of Verifiable Agent Conversation Record Data Items
The following table maps the data items of [VACR] to their
representation in this document.
+=========================+=======================+================+
| VACR data item | vCon representation | Extension |
| | | point |
+=========================+=======================+================+
| verifiable-agent-record | vCon Object | (core) |
+-------------------------+-----------------------+----------------+
| version | extensions parameter, | (core) |
| | "ai_agent" | |
+-------------------------+-----------------------+----------------+
| id | uuid parameter | (core) |
+-------------------------+-----------------------+----------------+
| created | created_at parameter | (core) |
+-------------------------+-----------------------+----------------+
| recording-agent | recording_agent | vcon-ext |
| | parameter | |
+-------------------------+-----------------------+----------------+
| vcs | vcs parameter | vcon-ext |
+-------------------------+-----------------------+----------------+
| file-attribution | file_attribution | vcon-ext |
| | parameter | |
+-------------------------+-----------------------+----------------+
| session-trace | Session Object | (core) |
+-------------------------+-----------------------+----------------+
| session-id | agent_session_id | session-ext |
| | parameter | |
+-------------------------+-----------------------+----------------+
| format | format parameter | session-ext |
+-------------------------+-----------------------+----------------+
Birkholz, et al. Expires 6 May 2027 [Page 57]
Internet-Draft vCon AI Agent Extension November 2026
| session-start | start parameter | (core) |
+-------------------------+-----------------------+----------------+
| session-end | duration parameter | (core) |
+-------------------------+-----------------------+----------------+
| environment | environment parameter | session-ext |
+-------------------------+-----------------------+----------------+
| agent-meta | agent_meta parameter | party-ext |
| | of the agent Party | |
| | Object | |
+-------------------------+-----------------------+----------------+
| entries | Dialog Objects and | (core) |
| | Event Objects of the | |
| | Session Object | |
+-------------------------+-----------------------+----------------+
| message-entry | "text" type Dialog | (core) |
| | Object | |
+-------------------------+-----------------------+----------------+
| message-entry model-id | model_id parameter | dialog-ext |
+-------------------------+-----------------------+----------------+
| token-usage | token_usage parameter | dialog-ext |
+-------------------------+-----------------------+----------------+
| tool-call-entry | "tool-call" type | event-type-ext |
| | Event Object | |
+-------------------------+-----------------------+----------------+
| tool-call-entry name | tool_name parameter | event-type-ext |
+-------------------------+-----------------------+----------------+
| tool-result-entry | "tool-result" type | event-type-ext |
| | Event Object | |
+-------------------------+-----------------------+----------------+
| reasoning-entry | "reasoning" type | event-type-ext |
| | Event Object | |
+-------------------------+-----------------------+----------------+
| event-entry | "system-event" type | event-type-ext |
| | Event Object | |
+-------------------------+-----------------------+----------------+
| timestamp | start (Dialog Object) | (core) |
| | or time (Event | |
| | Object) parameter | |
+-------------------------+-----------------------+----------------+
| id, parent-id | entry_id, parent_id | dialog-ext, |
| | parameters | event-type-ext |
+-------------------------+-----------------------+----------------+
| children | parent_id parameter | dialog-ext, |
| | of the nested entries | event-type-ext |
+-------------------------+-----------------------+----------------+
| abstract-timestamp | Date ([VCON-CORE]) | (core) |
+-------------------------+-----------------------+----------------+
| signed-agent-record | Appendix B | (none) |
Birkholz, et al. Expires 6 May 2027 [Page 58]
Internet-Draft vCon AI Agent Extension November 2026
+-------------------------+-----------------------+----------------+
| trace-metadata | Appendix B | (none) |
+-------------------------+-----------------------+----------------+
Table 25: Mapping of VACR Data Items
All other member names of [VACR] are converted from kebab-case to
snake_case, as required by [VCON-CORE]. The CDDL in Appendix B is
not part of the vCon schema and is used as in [VACR]. The media type
"application/agent-conversation", the CoAP Content-Format, and the
CBOR Tag requested by [VACR] are not needed, as a vCon uses the media
type "application/vcon".
Editor's Note: The Event Object of [VCON-CORE] requires the time
parameter, while the timestamp of an entry in [VACR] is optional.
The epoch milliseconds form of abstract-timestamp is not supported
by the Date type of [VCON-CORE].
Appendix D. Example vCons
The following (artificailly created) example is an unsigned vCon of a
short coding agent session. A user asks a question, the agent
reasons, reads a file using a tool, and answers.
{
"uuid": "01928e10-193e-8231-b9a2-079d3c1a8b5e",
"created_at": "2026-10-26T14:02:11Z",
"extensions": ["ai_agent"],
"recording_agent": {
"name": "example-recorder",
"version": "0.1.0"
},
"vcs": {
"type": "git",
"revision": "4f2a9c1",
"branch": "main",
"repository": "https://git.example.com/project.git"
},
"parties": [
{
"name": "Alice",
"type": "person"
},
{
"name": "Coding Agent",
"type": "bot",
"agent_meta": {
"model_id": "example-model-2026-10",
Birkholz, et al. Expires 6 May 2027 [Page 59]
Internet-Draft vCon AI Agent Extension November 2026
"model_provider": "example-provider",
"cli_name": "example-cli",
"cli_version": "1.2.0"
}
}
],
"sessions": [
{
"start": "2026-10-26T14:00:00Z",
"duration": 95,
"parties": [0, 1],
"dialog": [0, 1],
"agent_session_id": "8d1f6c2e-5a77-4c1e-9d4b-2f3a6b7c8d9e",
"format": "interactive",
"environment": {
"working_dir": "/home/alice/project",
"vcs": {
"type": "git",
"revision": "4f2a9c1",
"branch": "main"
}
}
}
],
"dialog": [
{
"type": "text",
"start": "2026-10-26T14:00:05Z",
"parties": 0,
"mediatype": "text/plain",
"encoding": "none",
"body": "What does the function compute_rate do?",
"entry_id": "e1"
},
{
"type": "text",
"start": "2026-10-26T14:01:30Z",
"parties": 1,
"mediatype": "text/plain",
"encoding": "none",
"body": "It computes the interest rate via a database lookup.",
"model_id": "example-model-2026-10",
"token_usage": {
"input": 1834,
"output": 212,
"total": 2046
},
"entry_id": "e5",
Birkholz, et al. Expires 6 May 2027 [Page 60]
Internet-Draft vCon AI Agent Extension November 2026
"parent_id": "e1"
}
],
"events": [
{
"type": "reasoning",
"time": "2026-10-26T14:00:07Z",
"session": 0,
"party": 1,
"content": "Read the source of compute_rate first.",
"entry_id": "e2",
"parent_id": "e1"
},
{
"type": "tool-call",
"time": "2026-10-26T14:00:09Z",
"session": 0,
"party": 1,
"tool_name": "Read",
"input": {"file_path": "src/rate.py"},
"call_id": "call-1",
"entry_id": "e3",
"parent_id": "e1"
},
{
"type": "tool-result",
"time": "2026-10-26T14:00:10Z",
"session": 0,
"party": 1,
"output": "def compute_rate(user): ...",
"call_id": "call-1",
"status": "success",
"is_error": false,
"entry_id": "e4",
"parent_id": "e3"
},
{
"type": "system-event",
"time": "2026-10-26T14:01:35Z",
"session": 0,
"party": 1,
"event_type": "session-end"
}
]
}
Figure 8: Example vCon with an Agent Session
Birkholz, et al. Expires 6 May 2027 [Page 61]
Internet-Draft vCon AI Agent Extension November 2026
Appendix E. CDDL for the ai_agent Extension
This appendix provides a [CDDL] definition for the parameters defined
by the ai_agent extension. The text in the preceding sections of
this document defines the normative definition of these parameters.
If the CDDL provided below differs from that in the text of the
preceding sections, the text is the definitive definition and the
CDDL below is to be considered incorrect.
This CDDL extends the vcon-ext, party-ext, session-ext, dialog-ext,
and event-type-ext sockets of the vCon CDDL of [VCON-CORE]. It is to
be used together with the vCon CDDL, which comes first so that its
first rule is the root of the combined definition. The fragments
below are the same ones that are included in the preceding sections,
in the order in which they are included by the main CDDL file of this
extension, which starts with the root rule "vcon-with-ai-agent =
vcon" and includes the vCon CDDL first.
$$vcon-ext //= ( recording_agent: recording_agent_type )
$$vcon-ext //= ( vcs: vcs_context_type )
$$vcon-ext //= ( file_attribution: file_attribution_type )
; Recording_Agent Object
recording_agent_type = {
name: tstr,
? version: tstr,
* tstr => any
}
; File_Attribution Object
file_attribution_type = {
files: [* file_type],
* $$file-attribution-ext
}
; File Object
file_type = {
; relative path from the repository root
path: tstr,
conversations: [* file_conversation_type],
* $$file-ext
}
; File_Conversation Object
file_conversation_type = {
? url: uri_type,
; default contributor for all ranges
? contributor: contributor_type,
Birkholz, et al. Expires 6 May 2027 [Page 62]
Internet-Draft vCon AI Agent Extension November 2026
ranges: [* range_type],
? related: [* resource_type],
* $$file-conversation-ext
}
; Range Object
range_type = {
; 1-indexed
start_line: uint,
; 1-indexed, inclusive
end_line: uint,
? content_hash: tstr,
; default: "sha-256"
? content_hash_alg: tstr,
; overrides the conversation level contributor
? contributor: contributor_type,
* $$range-ext
}
; Contributor Object
contributor_type = {
type: "human" / "ai" / "mixed" / "unknown",
? model_id: tstr,
* $$contributor-ext
}
; Resource Object
resource_type = {
type: tstr,
url: uri_type,
* $$resource-ext
}
; URI (RFC 3986)
uri_type = tstr .regexp
"(([^:/?#]+):)?(//([^/?#]*))?([^?#]*)(\\?([^#]*))?(#(.*))?"
$$party-ext //= ( agent_meta: agent_meta_type )
; Agent_Meta Object
agent_meta_type = {
; e.g., "claude-opus-4-5-20251101"
model_id: tstr,
; e.g., "anthropic", "google"
model_provider: tstr,
; all models (multi-model sessions)
? models: [* tstr],
; e.g., "claude-code", "gemini-cli"
Birkholz, et al. Expires 6 May 2027 [Page 63]
Internet-Draft vCon AI Agent Extension November 2026
? cli_name: tstr,
? cli_version: tstr,
* tstr => any
}
$$session-ext //= ( agent_session_id: tstr )
; "interactive" / "autonomous" / vendor specific
$$session-ext //= ( format: tstr )
$$session-ext //= ( environment: environment_type )
; Environment Object
environment_type = {
working_dir: tstr,
? vcs: vcs_context_type,
; sandbox mount paths
? sandboxes: [* tstr],
* tstr => any
}
$$dialog-ext //= ( model_id: tstr )
$$dialog-ext //= ( token_usage: token_usage_type )
$$dialog-ext //= ( entry_id: agent_entry_id_type )
$$dialog-ext //= ( parent_id: agent_entry_id_type )
; Token_Usage Object
token_usage_type = {
; input tokens
? input: uint,
; output tokens
? output: uint,
; cached input tokens
? cached: uint,
; reasoning or thinking tokens
? reasoning: uint,
; total tokens
? total: uint,
; monetary cost in US dollars
? cost: number,
* tstr => any
}
$$event-type-ext //= (event_tool_call_type,)
$$event-type-ext //= (event_tool_result_type,)
$$event-type-ext //= (event_reasoning_type,)
$$event-type-ext //= (event_system_event_type,)
; "tool-call" type Event Object
event_tool_call_type = (
Birkholz, et al. Expires 6 May 2027 [Page 64]
Internet-Draft vCon AI Agent Extension November 2026
type: "tool-call",
event_common_type,
; e.g., "Bash", "Edit", "Read"
tool_name: tstr,
; tool arguments
input: any,
; links call and result
? call_id: tstr,
agent_entry_reference_type,
* $$tool-call-event-ext
)
; "tool-result" type Event Object
event_tool_result_type = (
type: "tool-result",
event_common_type,
; tool output
output: any,
; links call and result
? call_id: tstr,
; "success" / "error" / "completed"
? status: tstr,
? is_error: bool,
agent_entry_reference_type,
* $$tool-result-event-ext
)
; "reasoning" type Event Object
event_reasoning_type = (
type: "reasoning",
event_common_type,
; plaintext reasoning or structured content
content: any,
; provider protected content
? encrypted: tstr,
; topic label
? subject: tstr,
agent_entry_reference_type,
* $$reasoning-event-ext
)
; "system-event" type Event Object
event_system_event_type = (
type: "system-event",
event_common_type,
; event classifier
event_type: tstr,
; event specific payload
Birkholz, et al. Expires 6 May 2027 [Page 65]
Internet-Draft vCon AI Agent Extension November 2026
? data: { * tstr => any },
agent_entry_reference_type,
* $$system-event-ext
)
vcs_context_type = {
; "git" / "jj" / "hg" / "svn"
type: tstr,
; commit SHA or change ID
? revision: tstr,
? branch: tstr,
; repository URL
? repository: tstr,
* tstr => any
}
; Reference to an entry of the native agent trace
agent_entry_reference_type = (
? entry_id: agent_entry_id_type,
? parent_id: agent_entry_id_type
)
; Per-entry unique reference within a session
agent_entry_id_type = tstr
Figure 9: Collated CDDL for the ai_agent Extension
Editor's Note: A JSON schema for the ai_agent extension will be
added.
Acknowledgments
Contributors
Tobias Heldt
XOR
Email: tobias@xor.tech
Thomas Fossati
NVIDIA
Email: tfossati@nvidia.com
Jaime Jimenez
Ericsson
Email: jaime@iki.fi
Birkholz, et al. Expires 6 May 2027 [Page 66]
Internet-Draft vCon AI Agent Extension November 2026
Jari Arkko
Ericsson
Email: jari.arkko@ericsson.com
Samuel Waymouth
AWS
Email: swaym@amazon.co.uk
Houda Labiod
Huawei Technologies France S.A.S.U.
Email: houda.labiod@huawei.com
Authors' Addresses
Henk Birkholz
Fraunhofer SIT
Email: henk.birkholz@ietf.contact
Daniel G Petrie
SIPez LLC
Email: dan.ietf@sipez.com
Thomas McCarthy-Howe
VCONIC
Email: ghostofbasho@gmail.com
Jonathan Rosenberg
jdrosen.net
Email: jdrosen@jdrosen.net
Orie Steele
Email: orie@or13.io
Birkholz, et al. Expires 6 May 2027 [Page 67]