.. _chapter-scope: Scope ===== Problem ------- Agents act across services and organizational boundaries. When an act is questioned afterwards, the party asking was usually not there, and what that party can establish depends on what was recorded at the moment of the act and on the properties of that record. FG-TIDA/themes#1 cites evaluation work by the UK AI Security Institute in which, where agents took unsanctioned actions, establishing what had happened "required an investigation rather than a lookup" (@OlehVasylenko2000, 10 August 2026). That evaluation ran with safeguards disabled, so the point is reconstruction, not the level of risk. The question is asked afterwards, but the requirement binds when the act happens (@OlehVasylenko2000, 15 August 2026). No persistent logical-agent identity is assumed. What the work item covers ------------------------- Requirements on the record, and on what a relying party reports from it: what an action-time record must be able to carry about the act (the execution that performed it and, where one exists, its supported relation to a logical agent; the delegated authority and declared scope; observed content, and a conformance verdict where one exists; a reference to the provenance of authority); the rejection classes a relying party reports (R1–R4); the properties of carried claims (R5–R7); absence and silence; lineage from representation to action; and identity continuity and effective state (R8). Out of scope ------------ Protocol specification: encodings, schemas, field names, media types, test vectors and expected-outcome files. How a relying party reaches a class, and how an execution, an identity, a state or a derivation relation is named. Model internals, and how any value in a record is computed. Replacing logging and audit practice. Legal liability. What a conformance verdict asserts and how it is checked (FG-TIDA/themes#6, task 3). Where authority originates and what may be conferred (FG-TIDA/themes#5). .. _chapter-requirements: Requirements ============ The clause text stands in 2 comments posted on FG-TIDA/themes#1: R1–R6 of 24 August 2026, and R4.4, R7 and R8 of 12 October 2026 (see :ref:`chapter-annexes`). This table summarises them; where it differs, the posted comments govern. Each row names who proposed the clause and, in brackets, who refined it. .. list-table:: Record-side requirements and their authors :header-rows: 1 :widths: 18 56 26 :class: longtable * - Clause - Requirement, in one line - Proposed (refined) by * - R1 Separability - A relying party that rejects reports the class of the rejection; a bare accept or reject is not enough. - @tyche-dev; asked for as a requirement by @OlehVasylenko2000 * - R2 The reported reason must be true - A correct rejection with a misattributed cause does not conform. - @tyche-dev * - R3 Tier 1 classes - Replayed, expired, malformed, out of mandate: each decidable from the record and the relying party's own state, and distinguishable from the others. - @tyche-dev (@AntoBud, @OlehVasylenko2000) * - R4 Tier 2: standing withdrawn - Every tier 1 check passes, but the grantor would not confer the mandate at the moment of the act. The record declares whether the grantor's continued standing is checkable (R4.1); a relying party records which moment it used (R4.2); neither moment is mandated, and the choice is visible (R4.3). - the class @AntoBud; the 2 tiers @tyche-dev * - R4.4 - A carried standing answer is a claim, and it carries what R7 requires. - @AntoBud * - R5 - Negative and null assertions are claims, subject to R1–R4 as a positive claim is. - @cgorlla; the null case @OlehVasylenko2000 (@zhaoxinghua09-cell) * - R6 - Freshness is a relation, not a property: the record carries the moment its assertion was true and the means, if any, of rechecking it. - @AntoBud * - R7 - Any carried claim needs attribution (R7.1), a stated moment (R7.2) and tamper-evidence that a party with no relationship to the producer can check (R7.3). R7 does not make a claim true. - the recursion @OlehVasylenko2000; the names follow @OlehVasylenko2000 and @danyka-icam * - Absence and silence - The record keeps 3 states apart: a reference was established; a bounded search, with its scope and author, returned nothing; nothing was recorded. - @pdixon-wpf and @OlehVasylenko2000, on FG-TIDA/themes#9 * - LR1–LR5 Lineage - Descent, transformed descent, independent reconstruction, examined-but-unresolved and record silence stay distinguishable; similarity and conformance never stand in for lineage, and missing evidence never yields NO. - @danyka-icam (@OlehVasylenko2000) * - R8 - A later relying party can establish the act, the execution that performed it, that execution's supported relation to a logical agent where one exists, and the effective state that governed the act at that time, not a later state (R8.1–R8.5). - @syedofc; R8.3 @OlehVasylenko2000; R8.5 @pdixon-wpf * - 3 relations - The execution, its supported relation to a logical agent, and the principal or harness context that instantiated it stay distinguishable; instantiation alone establishes no authorization. - @syedofc; supported by @OlehVasylenko2000 Questions carried open ---------------------- The work item carries the open questions of the theme 1 charter (PR #46, version 0.5), with the positions recorded under them: the evidence basis of a carried claim (1); whether the record states which checks were in the path (2); what attribution binds to (3); "and if so where" in R4.1 (5); a third UNCLEAR lineage state (6); multi-step delegation, the boundary with law and policy, and record-keeping obligations (7); inputs from FG-TIDA/themes#5, #36 and PR #28 (9); reconstructable or re-computable (10); whether a deployment can meet R8 (11); and an asserted or witnessed moment (12). On the last: R7.3 lets a party check that a record is unaltered, but not when it was written (@Abracadabrastartup, 10 November 2026). @OlehVasylenko2000 suggested that D1 first keep 3 moments apart: the moment of the act; the moment a producer states for a claim; and an independently bounded moment, which shows only that a record existed no later than a given time and is not the freshness relation of R6 (12 November 2026). If the proposed R7.2a becomes text, it is a sub-clause of R7.2, and R7 keeps 3 properties. .. _chapter-relation: Relation to other work ====================== * **WG2, Identity, Behaviour and Enforcement for AI Agents (IBEA).** The WG2 charter places "the record that carries a decision, and the properties any carried claim needs" under FG-TIDA/themes#1 (Out of Scope). For the texts that describe this boundary, @OlehVasylenko2000 proposed one line (PR #46, 12 November 2026), which this work item takes: "WG2 specifies what a decision is and what its outcome means, including what the decision record states; #1 specifies the record-side requirements that record must meet and the properties of the claims it carries (R1–R8)." * **WG1, Continuous and Dynamic Remote Attestation for Agentic AI (CDRA).** WG2 examines with WG1 whether a third party can later check an enforcement decision. That check also rests on R7.3: tamper-evidence that a party with no relationship to the producer can check. * **FG-TIDA/themes#5, Provenance of Authority: mutual, not sequential.** Provenance determines what a record can truthfully say about authority; the properties of the record determine whether it survives a party who was not there. * **FG-TIDA/themes#6, task 3.** What a conformance verdict asserts and how a verifier checks it. Task 3 takes its tier 1 and tier 2 rejection classes from R3 and R4 unchanged, and #6 "will reference" R7 rather than restate it (@cgorlla, 14 October 2026). * **FG-TIDA/themes#27 and PR #28 (verifier-side requirements in regulated workflows).** Of the change-event record it proposes, its drafter wrote that the record-side requirements in themes#1 "define where this record belongs; this item specifies what the verifier must be able to read from it" (@zhaoxinghua09-cell, PR #28, 15 October 2026). * **Verifier-side interface to the record requirements (linked contribution record).** Four clauses, VR1 to VR4, on the report a verifier makes: each check as a pair of who performed it and against what; the moment evaluated against; what a pass claims and within what boundary; and the report as a claim the next party assesses. The record "adds nothing to R1–R8 and alters nothing in them", and it stands beside this document, under its author's name (@zhaoxinghua09-cell, PR #46, 16 November 2026): https://github.com/LGD-Eval/evaluation-protocol/blob/f66c4658fd1c99b1e6fef99b99eed3865dc8e45f/FG-TIDA-AAR-VERIFIER-SIDE-INTERFACE.md. Whether the four clauses move into this work item is for the group to decide. * **WG0, Use Cases and Terminology.** Use cases #7 (@syedofc) and #12 (@danyka-icam) supply the cases for D2. * **Format work.** Encodings and profiles for carried claims belong to the bodies that specify formats. The thread named the Internet Engineering Task Force (IETF) Remote Attestation Procedures (RATS) architecture and the Entity Attestation Token (EAT) as relevant adjacent work; this assigns no standards body exclusive responsibility for future formats (@OlehVasylenko2000, PR #46, 26 October 2026). * **ITU-T Study Group 17 (SG17).** "Work under this theme complements ongoing SG17 work without duplicating it. Questions of possible overlap are resolved by the FG Chair and the SG17 Chair on a case-by-case basis, in accordance with clause 2 of the Terms of Reference." (wording by @OlehVasylenko2000, 16 September 2026) .. _chapter-programme: Programme of work ================= Deliverables ------------ * **D1 Record-side requirements.** One consolidated text of R1–R8, the lineage requirements, the absence states and the 3 relations, each clause under its author's name. Requirements only: no encodings, schemas, protocols, identifiers, identity-binding mechanisms or conformance vectors. * **D2 Cases.** A short mapping of the lineage cases LR1–LR5 and use cases #7 and #12 to the requirements they test. D2 fixes no conformance vectors. Placement --------- If the Focus Group places the work in an existing working group, or in one formed in Paris, D1 and D2 move into that group's deliverables, and no competing charter is opened: "the requirements travel, and the drafting responsibility travels with them" (@OlehVasylenko2000, 1 September 2026). Editor ------ Anton Sokolov offers to serve as editor, as offered on FG-TIDA/themes#1 on 31 August 2026 ("willing to serve as editor or co-editor, subject to the drafting timetable it adopts"), and asks for at least 1 co-editor for D1. Who co-edits, and the cadence, are for the group to set. Proposed timetable ------------------ Paris, 30 November – 1 December 2026: decision on the work item and its placement. Before the 2nd meeting (Geneva, 26–27 January 2027): a consolidated D1 draft for review. The editor attends in Paris on 30 November and can present the item that day. Attribution ----------- Clause-level attribution stays traceable in D1, where every clause stands under its author's name, and in the linked contribution record (see :ref:`chapter-annexes`): the theme 1 charter, which records who proposed and refined each clause and keeps the open questions, and the 2 requirement comments on themes#1, which hold the clause text. .. _chapter-annexes: Annex: source record ==================== * FG-TIDA/themes#1: https://github.com/FG-TIDA/themes/issues/1 * Requirement text R1–R6, 24 August 2026: https://github.com/FG-TIDA/themes/issues/1#issuecomment-5399524599 * Requirement text R4.4, R7 and R8, 12 October 2026: https://github.com/FG-TIDA/themes/issues/1#issuecomment-6127433298 * Theme 1 charter, version 0.5 (PR #46): https://github.com/FG-TIDA/themes/pull/46 References ============ [ITU-T A.7] Recommendation ITU-T A.7, *Focus groups: Establishment and working procedures*. [FG-TIDA ToR] FG-TIDA Terms of Reference, https://www.itu.int/en/ITU-T/focusgroups/tida/Pages/ToR.aspx [WG2 charter] FG-TIDA WG2, Identity, Behaviour and Enforcement for AI Agents (IBEA), charter, https://github.com/fg-tida-wg2/wg2-ibea-charter [Use cases] FG-TIDA WG0, use cases #7 and #12, https://github.com/fg-tida-wg0/use-cases/issues/7 https://github.com/fg-tida-wg0/use-cases/issues/12 .. raw:: latex \pagestyle{plain} \thispagestyle{plain} %% Suppress all TOC entries for the overview chapter (sections, subsections, %% Copyright addcontentsline) — restored just before \sphinxtableofcontents %% so chapters from Introduction onward appear in the TOC normally. \let\osiaorigaddcontentsline\addcontentsline \renewcommand{\addcontentsline}[3]{} .. _chapter-overview: Accountability and Attribution After an Agent Has Acted ======================================================= .. only:: html **Record-side requirements (AAR-REQ). Proposal for a new work item** .. raw:: latex \begin{center}\textbf{Record-side requirements (AAR-REQ). Proposal for a new work item}\end{center} Abstract -------- When an act by an agent is questioned later, the party asking was usually not present, and establishing what happened can require "an investigation rather than a lookup" (FG-TIDA/themes#1). This contribution proposes a work item that states what an action-time record must be able to carry, and what properties every claim it carries needs, so that such a party can establish the execution that performed the act and, where one exists, its supported relation to a logical agent; under whose delegated authority; within what declared scope; and that the record is unaltered. It consolidates requirements R1–R8, the lineage requirements and the absence states developed in the open thread on themes#1 since August 2026, each clause under its author's name, and maps them to cases. It states requirements only and prescribes no encodings, schemas, protocols or identifiers. History ------- .. list-table:: :header-rows: 1 :widths: 10 26 18 14 32 * - Version - Document - Approval - Focus Group - Working Group * - 0.1 - AAR-REQ: proposal for a new work item - — - TIDA - Incubation (proposed) * - 0.2 - Co-author added; status and copyright wording corrected - — - TIDA - Incubation (proposed) * - 0.3 - Two co-authors and a contributor added; a linked contribution record listed - — - TIDA - Incubation (proposed) To access this Document as a PDF, open the published documentation site in your web browser and click the PDF download link, or navigate directly to https://tyche-dev.github.io/tida-aar-req/specification.pdf. Authors ------- .. list-table:: :header-rows: 1 :widths: 30 40 30 * - Name - Affiliation - Contact * - Anton Sokolov (editor) - Tyche Institute MTÜ, Estonia - anton.sokolov@tyche.institute; GitHub @tyche-dev * - Syed Amir Hamza - Forkit AI - GitHub @syedofc * - Nika Novak - ICAM Research Programme - GitHub @danyka-icam * - Xinghua Zhao - Independent expert - GitHub @zhaoxinghua09-cell Contributors. The requirements come from the open thread on FG-TIDA/themes#1. Its originating proposer is Oleh Vasylenko (@OlehVasylenko2000), Management Company Elixir LLC (Ukraine), who takes part as proposer and reviewer. Co-author Nika Novak (@danyka-icam) co-drafted the lineage requirements and cases and contributed use case #12. Co-author Syed Amir Hamza (@syedofc) co-drafted the identity and state continuity requirements (R8) and contributed use case #7. Co-author Xinghua Zhao (@zhaoxinghua09-cell) is named for the null-case refinement on R5 and for the linked contribution record on the verifier side (:ref:`chapter-relation`). @Abracadabrastartup (DEUSPROOF) raised the point recorded under open question 12, an asserted or witnessed moment (a possible R7.2a). Every clause in :ref:`chapter-requirements` names its author. Keywords -------- Agentic AI; accountability; attribution; action-time record; carried claim; rejection class; delegated authority; lineage; tamper-evidence. .. raw:: latex \newpage \osiacenternextsection Foreword -------- Agentic AI systems now act as autonomous entities capable of reasoning, planning, and executing tasks across services, infrastructures, and organizational boundaries. This fundamentally changes the nature of identity in digital systems: identity is no longer limited to human users and static machine instances but must also encompass autonomous, dynamic, and goal-oriented agents. And identity alone is insufficient — the core question extends beyond "what an entity is" to "whether, and under what conditions, that entity should be trusted to act". The Focus Group on Trust and Identity for humans and agentic AI (FG-TIDA) addresses trust management and interoperable digital identity infrastructure for humans and for agentic AI, supporting the development of secure, trustworthy digital ecosystems in which humans and agentic AI can safely interact and collaborate. FG-TIDA was established under Recommendation ITU-T A.7 (Focus groups: Establishment and working procedures), with ITU-T Study Group 17 as its parent group. Further details are available in the group's terms of reference. For more news and updates about the TIDA Community, including its supporter organizations, please refer to the community's published resources. .. raw:: latex \osiacenternextsection Note ---- This document is a proposal for a new work item, submitted to FG-TIDA Incubation on 15 November 2026 as an input to the first FG-TIDA meeting (Paris, 30 November – 1 December 2026). The TIDA DataTracker records it as contribution FG-TIDA-INCUBATION-003, state Submitted. It is not adopted Focus Group text. The requirement text it summarises stands in the posted comments listed in :ref:`chapter-annexes`. .. Intellectual Property Rights — section commented out. To restore it, remove the leading ".." and the indentation below, and re-add a ".. raw:: latex" block with "\osiacenternextsection" before the heading (to centre it on the PDF). Intellectual Property Rights ---------------------------- ITU draws attention to the possibility that the practice or implementation of this Recommendation may involve the use of a claimed Intellectual Property Right. ITU takes no position concerning the evidence, validity or applicability of claimed Intellectual Property Rights, whether asserted by ITU members or others outside of the Recommendation development process. As of the date of approval of this Recommendation, ITU had not received notice of intellectual property, protected by patents/software copyrights, which may be required to implement this Recommendation. However, implementers are cautioned that this may not represent the latest information and are therefore strongly urged to consult the appropriate ITU-T databases available via the ITU-T website at http://www.itu.int/ITU-T/ipr/. Implementers should also be aware that the organization that originated the technically equivalent document listed in the Bibliography may have received notices of intellectual property required for the implementation of this Recommendation. .. raw:: latex %% Copyright notice commented out. To restore, uncomment the lines below. %% \vspace{4em} %% \begin{center}\textcopyright{} ITU 2024\end{center} %% %% \medskip %% All rights reserved. No part of this publication may be reproduced, by any means whatsoever, without the %% prior written permission of ITU. %% Restore \addcontentsline so the TOC below (and Chapter 1 onward) populate. \let\addcontentsline\osiaorigaddcontentsline \sphinxtableofcontents \osiamainmatter .. toctree:: :maxdepth: 3 :numbered: :hidden: chapters/01_scope chapters/02_requirements chapters/03_relation_to_other_work chapters/04_programme_of_work chapters/05_annex/index chapters/07_references