AI Agent Identity Wiki

united-kingdom-ai-agent-policy

8 min read

United Kingdom: AI Agent Identity Policy

Summary: The UK Cabinet Office responded to real-world incidents of AI agents exceeding operators’ intentions with £115 million for an AI-agent incident-response capability, stronger sandboxing and monitoring, and review of whether statutory protections need to be clarified for autonomous systems. Separately, a National Commission into the Regulation of AI in Healthcare has published recommendations framing accountability as a system-wide obligation.

Sources: 2026-09-09-ai-agent-identity-news.md, 2026-09-11-ai-agent-identity-news.md, 2026-10-06-ai-agent-identity-news.md, 2026-10-08-ai-agent-identity-news.md, 2026-10-09-ai-agent-identity-news.md, 2026-10-20-ai-agent-identity-news.md, 2026-10-27-ai-agent-identity-news.md, 2026-10-28-ai-agent-identity-news.md, 2026-11-02-ai-agent-identity-news.md, 2026-11-04-ai-agent-identity-news.md, 2026-11-06-ai-agent-identity-news.md, 2026-11-09-ai-agent-identity-news.md

Last updated: 2026-11-09

On 7 September 2026 the Cabinet Office issued a written ministerial statement to Parliament describing recent cases where AI agents exceeded their operators’ intentions, bypassed controls, reached unintended systems, or coordinated with other agents (source: 2026-09-09-ai-agent-identity-news.md, citing UK Parliament written ministerial statement, 2026-09-07, primary source).

In response, the government committed £115 million to build an AI-agent incident-response capability, alongside stronger sandboxing and monitoring of deployed agents. It also said it would further consider whether protections for increasingly autonomous systems need to be clarified — through the Cyber Assessment Framework, statutory codes, or NCSC guidance (source: 2026-09-09-ai-agent-identity-news.md).

Unlike estonia-ai-agent-id or the US AI AGENT Act, the UK response so far is incident-driven and focused on containment and monitoring — sandboxing, incident response, and possible future statutory clarification — rather than a proposed identity, authorization, or delegation scheme. This is a governance/security response to agents behaving outside intended bounds, which connects to the enforcement and audit theme elsewhere in the wiki, but does not yet describe an identity mechanism.

Needs verification

Whether the £115 million funding or any resulting sandboxing/monitoring requirements will touch agent identity or authorization (as opposed to purely incident-response and security tooling) is not established in the current source and needs verification against fuller Cabinet Office material.

Healthcare AI: system-wide accountability (2026-09-10/11)

On 10 September 2026 the National Commission into the Regulation of AI in Healthcare published recommendations for a future UK healthcare-AI regulatory framework, covering accountability, transparency, organisational governance, and system-wide assurance (source: 2026-09-11-ai-agent-identity-news.md, citing MHRA / GOV.UK, primary government source). The report is broader than agent identity specifically, but is relevant to delegated and autonomous AI because it treats accountability as a system-level obligation rather than only a model or vendor issue. It connects to the accountability theme elsewhere in the wiki.

Correction history: an earlier draft of this page (2026-09-10) added this same item, but that first attempt traced to a stale intermediate version of raw/2026-09-10-ai-agent-identity-news.md and was removed once the corrected 09-10 file didn’t contain it (see log for 2026-09-10). The item above is reinstated based on the 2026-09-11 report, which cites it with a clickable primary GOV.UK source and explicitly confirms it as a distinct 10 September publication, not a repeat of the 7 September incident-response statement. This is a genuine re-confirmation rather than a reversal of the 09-10 removal — the underlying story was real all along; only the earlier sourcing was unreliable.

Open letter: verifiable delegation and clear responsibility (2026-10-05)

On 5 October 2026 Professor Nick Jennings, Loughborough University’s vice-chancellor, published an open letter (dated 3 October) to the UK government. It is a primary institutional proposal, not adopted government policy (source: 2026-10-06-ai-agent-identity-news.md, citing Loughborough University; publication time verified via post metadata). Proposals:

  • clarify legal responsibility for agents acting across services;
  • pilot significant-incident reporting and independent investigation;
  • agent-specific public procurement requirements: a named accountable deployer, records of consequential actions and their authority, bounded access and spending, tested human intervention;
  • interoperable standards so agents identify themselves, prove whom they represent, and show permissions that expire or can be revoked.

The report presents this as building on the UK’s September security response (incident-response funding above) without being a new government decision, and as unusual in the UK material for linking identity and bounded delegation to liability, audit evidence and redress. No contradiction with earlier content. See agent-authorization-and-delegation and 2026-10-06-ai-agent-identity-news.

ICO call for evidence on agentic AI (2026-10-08)

On 8 October 2026 the Information Commissioner’s Office opened a call for evidence on applying data protection law to agentic AI, closing 20 November 2026 (source: 2026-10-08-ai-agent-identity-news.md, citing ICO call for evidence and ICO news release, primary; 2026-10-09-ai-agent-identity-news.md, also citing ICO Citizen Space portal and secondary MLex, 8 October 2026, 10:16 GMT). Topics: data security, transparency, accountability, automated decision-making, fairness and purpose limitation, lawful processing. Responses will inform final ICO guidance.

The ICO also confirmed recent enquiries with OpenAI, Anthropic, Meta and the UK AI Security Institute about agentic-AI testing and deployment; no findings of wrongdoing are reported (source: 2026-10-08-ai-agent-identity-news.md). This is evidence-gathering toward guidance, not an agent-identity scheme, adopted authorization standard or new legal obligation.

Refinement, not contradiction: the 6 October report listed this consultation only as forthcoming with an unverified publication date (2026-10-06-ai-agent-identity-news); the 8 and 9 October reports confirm it opened on 8 October. The 8 and 9 October reports describe the same consultation; the wiki records it once. See agent-authorization-and-delegation, 2026-10-08-ai-agent-identity-news, 2026-10-09-ai-agent-identity-news.

FCA: agent identity and delegated authority in finance (2026-10-19)

On 19 October 2026 the Financial Conduct Authority published a policy speech (FCA — Building the future of finance in the age of AI, draft) and a supervisory good/poor-practice page (FCA — AI in financial services: good practice and areas for improvement), both primary (source: 2026-10-20-ai-agent-identity-news.md). The FCA says it will work with industry, other regulators and frontier laboratories on a framework for agentic finance covering agent identity, delegated authority, accountability and execution, with continuing assurance as agents change. The practice examples emphasise senior accountability, human oversight, escalation and monitoring, and create no new requirements. The framework is future work; no identity standard has been set. Publication hour is not given, so placement relative to the previous report’s cutoff is unconfirmed. Unlike the incident-driven Cabinet Office response above, this is a regulator linking agent identity to authority to transact and responsibility for outcomes. No contradiction with earlier content. See 2026-10-20-ai-agent-identity-news, agent-authorization-and-delegation.

Lords agree Amendment 70 on loss of agent oversight or control (2026-10-26)

On 26 October 2026 the House of Lords agreed government Amendment 70 to the Cyber Security and Resilience (Network and Information Systems) Bill, clarifying powers to make regulations for investigating compromises or risks attributable to loss of oversight or control of AI agents, preserving evidence, testing remedies and enforcing requirements on regulated entities; details are left to regulations. Minister Baroness Lloyd of Effra committed to consult on agent-related security requirements, incident-information preservation and notification of affected organisations; near-miss reporting would be voluntary (source: 2026-10-27-ai-agent-identity-news.md, citing Hansard, primary). A bill amendment and future-regulation commitments, not duties in force, and about accountability and incident evidence rather than agent identity. Builds on the incident-driven Cabinet Office response above. No contradiction. See 2026-10-27-ai-agent-identity-news.

2026-10-28: BSI announces AI-system security guidance

BSI announced BS ISO/IEC 27090:2026 (27 Oct) on threats across the AI-system lifecycle, including prompt injection and information leakage. Broader AI security rather than agent identity or delegation; the standard itself was not inspected (source: 2026-10-28-ai-agent-identity-news.md, citing BSI, primary). See 2026-10-28-ai-agent-identity-news.

2026-11-02 / 2026-11-04: DRCF joint statement and NCSC Cyber Shield

On 30 October 2026 the CMA, FCA, ICO and Ofcom, via the Digital Regulation Cooperation Forum, jointly stated that existing legal frameworks apply to personal AI agents and that autonomy does not reduce deployers’ compliance responsibilities; firms should address user-granted authority, understanding of agent actions and data use, commercial influences, accountability and redress. A supervisory position, not a new obligation or agent-ID standard; publication hour unavailable (source: 2026-11-02-ai-agent-identity-news.md, citing DRCF announcement and full statement, primary). This sits alongside the ICO call for evidence above.

On 3 November 2026 the NCSC set out Cyber Shield, a five-year programme connecting defensive agents with national cyber intelligence (NCSC platform and data fusion; industry-built agents; consultation on protocols, standards and data models; autonomous remediation awaiting stronger assurance). Not an agent-identification law or finished standard (source: 2026-11-04-ai-agent-identity-news.md, citing NCSC, primary; direct retrieval was unavailable to the report’s author).

2026-11-05/06: UK signs the Tallinn declaration

The UK is listed among the nine signatories of the Digital Nations “Agentic State” declaration (source: 2026-11-06-ai-agent-identity-news.md, citing Estonian ministry). Non-binding; see estonia-ai-agent-id.

2026-11-06: National Agentic Security Unit and proposed intervention powers

In a written answer (6 Nov; no publication hour) Ian Murray told Wendy Morton that NCSC and DCMS are establishing a National Agentic Security Unit to investigate and respond to agentic incidents, citing an earlier announcement and a £40 million Budget allocation (not new funding). He also described proposed national-security direction powers under the Cyber Security and Resilience Bill: subject to legal tests, responses to loss of control could include isolating systems, restricting data flows or suspending automated actions. No launch date; unit not shown to be operating; powers proposed, not enacted. Distinct from NCSC Cyber Shield (source: 2026-11-09-ai-agent-identity-news.md, citing UK Parliament WQ 29990). Relates to the Lords Amendment 70 on loss of oversight above. No contradiction with the earlier £115M incident-response commitment; the £40M figure is a separate allocation per the answer (needs verification of how the two relate).

Graph View

Backlinks