2026-10-27 AI Agent Identity News
Summary: Dated digest of raw/2026-10-27-ai-agent-identity-news.md (Tuesday). Estonia: ERR coverage of expert criticism (Cybernetica) of the proposed agent registry and its consultation status. Other countries: UK Lords agree Amendment 70 on loss of oversight or control of AI agents; FTC seeks Anthropic executives’ testimony; Australia’s National AI Centre agentic-AI guidance; ARI reporting proposal; IBM Agent Identity generally available. No research (Tuesday).
Sources: 2026-10-27-ai-agent-identity-news.md
Last updated: 2026-10-27
Section 1 — Estonia
- Cybernetica interview (ERR, 26 Oct, 17:27 Helsinki; secondary, expert interview): Cybernetica development lead Arne Ansper argues the proposed voluntary registry needs substantial redesign because legal work is proceeding before the technical mechanism is defined. He separates legal personhood from technical identity, says numbering agents is insufficient without authentication and a way to limit agent authority, and warns the design affects Estonia’s wider digital-identity infrastructure. These are the interviewee’s assessments, not a regulator’s ruling (source: 2026-10-27-ai-agent-identity-news.md, citing ERR — Cybernetica arendusjuht).
- Consultation status (ERR, 26 Oct; secondary): the report says the minister explained that the initial arrangement would allow only private individuals, one registered agent per person; and, per the English account (09:16 Helsinki), no ministry approved the draft by the deadline, although the Government Office did. Revisions are planned before a hoped-for early-November parliamentary submission. Not passage of the bill (source: 2026-10-27-ai-agent-identity-news.md, citing ERR — Experts consider the state’s plan dangerous and ERR News — English). A 25 Oct ERR radio summary is context only.
- The MKM agendikood FAQ (primary, updated 23 Oct) remains the reference for distinguishing an agent code from the isikukood (source: 2026-10-27-ai-agent-identity-news.md). See estonia-ai-agent-id.
Section 2 — Other countries
- United Kingdom: the House of Lords agreed government Amendment 70 to the Cyber Security and Resilience (Network and Information Systems) Bill on 26 Oct, clarifying regulation-making powers to investigate compromises or risks attributable to loss of agent oversight or control, preserve evidence, test remedies and enforce requirements on regulated entities. Baroness Lloyd of Effra committed to consultation on agent security requirements, incident-information preservation and notification of affected organisations; near-miss reporting would be voluntary. A bill amendment and commitments, not duties in force; not an agent-ID scheme (source: 2026-10-27-ai-agent-identity-news.md, citing Hansard, 26 Oct, primary). See united-kingdom-ai-agent-policy.
- United States — FTC / Anthropic: the FTC confirmed to The Hill that it demanded testimony from Anthropic executives in its consumer-harm investigation, following earlier document demands; no confirmation of equivalent demands for OpenAI. Anthropic’s 26 Oct SEC filing acknowledges an October civil investigative demand (consumer protection, privacy, data protection, product safety/security), corroborating the document demand; the testimony detail rests on the news report. No wrongdoing established (source: 2026-10-27-ai-agent-identity-news.md, citing The Hill via AOL, secondary; Anthropic S-1, primary). See united-states-ai-agent-policy.
- United States — ARI proposal: Americans for Responsible Innovation proposes confidential quarterly filings to a new Commerce Department office on automated AI research, including the share of AI-generated inputs and decisions reviewed by humans, with reporting within four business days of material shifts such as fully automating a research loop. An advocacy recommendation, not adopted (source: 2026-10-27-ai-agent-identity-news.md, citing ARI, primary; Nextgov/FCW, secondary).
- United States — IBM Agent Identity GA: IBM says Agent Identity is generally available (previously September private preview): central agent inventory with ownership and lifecycle state, on-behalf-of delegation across token exchanges, short-lived task-scoped tokens, runtime policy checks and audit trails; IBM Verify and HashiCorp Vault supply identity/authorization and credential protection. Vendor claims, not independently tested; publication hour unverified (source: 2026-10-27-ai-agent-identity-news.md, citing IBM, primary corporate).
- Australia: the National AI Centre published guidance, Understanding and using agentic AI, covering limits, approvals, oversight, tool/provider assessment, implementation and review. Only the listing and scope were verified; the guide itself was not inspected, so no detailed controls are attributed; publication hour unavailable (source: 2026-10-27-ai-agent-identity-news.md, primary). See australia-ai-agent-policy.
Section 3 — Research
Omitted (Tuesday).
Contradictions
None. Refinement only: the registry is described as voluntary, initially for private individuals with one agent per person, and the draft lacked ministry approval at the deadline; this adds detail to the “proposed registry” already recorded and does not overturn the 2026-10-25 agendikood position.