|
Work item:
|
XSTR.llmp-tip
|
|
Subject/title:
|
Technical Report: LLM token-level inference provenance
|
|
Status:
|
Under study
|
|
Approval process:
|
Agreement
|
|
Type of work item:
|
Technical report
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
2028-12 (Medium priority)
|
|
Liaison:
|
ITU-T SG21, AMAS intiative
|
|
Supporting members:
|
Management Company Elixir LLC; Broadcom Europe Ltd
|
|
Summary:
|
XSTR.llmp-tip provides specification of token-level inference provenance mechanisms for large language model outputs, operationalizing the inference-provenance layer underlying transparency obligations for generative AI.
The Technical Reportdefines token-level provenance primitives, inference provenance graph structures, token genesis records, and inference attestation procedures supporting verifiable presentation of token-level provenance to relying parties. The work item complements adjacent work on large language model supply chain controls (X.scm-llm) and content controls (X.LLMCC), and is accompanied by a formal liaison statement to SG21 (AMAS initiative) addressing complementary asset-level media provenance. The work item contributes to the Q16/17 work programme on artificial intelligence security as adopted at the SG17 4th Plenary (Geneva, 9 April 2026). Regulatory mapping anchors: EU AI Act Article 50 (transparency obligations for providers and deployers of certain AI systems, applicable from 2 August 2026); C2PA Technical Specification version 2.2 (Coalition for Content Provenance and Authenticity, Joint Development Foundation project, May 2025) as adjacent standard for asset-level provenance; W3C PROV-DM (W3C Recommendation, 30 April 2013) as upstream data model; ISO/IEC 42001 Annex A.7 (Data for AI systems).
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2026-06-10 09:29:16
|
|
Last update:
2026-10-15 17:28:46
|
