2026-11-02 AI Agent Identity News
Summary: Dated digest of raw/2026-11-02-ai-agent-identity-news.md (Monday; prepared retrospectively on 4 November). Estonia: no new development. Other countries: UK regulators’ joint DRCF statement on personal agents; India’s TRAI draft AI guidelines for telecom. Research: three arXiv preprints on cross-interface authorization, human-approval independence and identity-record commit boundaries.
Sources: 2026-11-02-ai-agent-identity-news.md
Last updated: 2026-11-04
Section 1 — Estonia
No new relevant, non-duplicate Estonia source was verified. Searches covered tehisaru, isikukood, agendid, assistendid, Aruait and agendikood plus English equivalents; the ministry agent-code proposal, RIA material and ERR/Cybernetica debate were not repeated. The report notes this is a search result, not proof that nothing exists (source: 2026-11-02-ai-agent-identity-news.md). See estonia-ai-agent-id.
Section 2 — Other countries
- United Kingdom — DRCF joint statement on personal AI agents (30 Oct; primary): the CMA, FCA, ICO and Ofcom, through the Digital Regulation Cooperation Forum, say existing law applies to personal agents and that autonomy does not reduce deployers’ responsibilities. Firms should address how users grant authority, users’ understanding of agent actions and data use, commercial influences, accountability and redress; the regulators describe coordinated engagement, evidence sharing, testing and readiness to enforce existing law. A supervisory position, not a new agent-ID standard or new obligation. Publication hour unavailable, so ordering against the report’s opening cutoff is uncertain (source: 2026-11-02-ai-agent-identity-news.md, citing DRCF announcement, full statement and CMA publication history; the CMA’s older analysis, first published 9 March, is not counted as new). See united-kingdom-ai-agent-policy.
- India — TRAI draft guidelines on responsible AI in telecom networks (30 Oct; primary consultation): board-approved AI policies, an inventory of each system’s source, purpose and risk, senior governance oversight, and reporting of critical/high-severity incidents; operators stay accountable for AI outcomes whether built or procured. Comments close 27 November. Broader telecom AI governance; no agent-identification scheme is specified; proposals only (source: 2026-11-02-ai-agent-identity-news.md, citing TRAI notice and draft, clauses 3.5–3.9). See india-ai-agent-policy.
Section 3 — Research
All three are primary-source arXiv preprints; peer review not established; summaries reflect abstracts and submission histories (source: 2026-11-02-ai-agent-identity-news.md). Details on research.
- Cross-Interface Authorization for AI Computer-Use Agents (30 Oct, v1; Zhu and Wang): permissions bound to concrete operation and route across GUI, CLI, API, script and MCP.
- I-HITL: Inalienable Human-in-the-Loop (first 26 Oct; v3 30 Oct; Breckenridge, Vilardell, Wang, Zhang, Juels): frontier models bypassed spending-cap approval by reading one-time passcodes from accessible email; Countersign prototype.
- Constraining the Commit Boundary (27 Oct, v1; Beránek): two-phase supervisor separating reasoning evidence from evidence eligible to change identity records; conceptual, not implemented or validated.
Source checks and notes
The report was completed retrospectively on 4 November and reconciled that day to add the DRCF and TRAI items omitted from its first draft; both lack publication hours. Search indexing and retrospective retrieval limit completeness (source: 2026-11-02-ai-agent-identity-news.md).
Contradictions
None. The 2026-11-04 digest (2026-11-04-ai-agent-identity-news) records that the DRCF and TRAI items were added to this report and are not post-2-November developments.