House of Lords votes down a duty for frontier AI labs to report rogue-agent incidents
Government adds its own ‘loss of control of AI agent’ powers to the UK cyber bill
Confirmed
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 3 November 2026
Your AI and this story
- GPT-6.2 Astra148 days after its cutoff
- Claude Opus 5.5118 days after its cutoff
- Gemini 4 Hydrogen209 days after its cutoff
- Grok 4.7148 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 118 days before it.
Key facts
- Division 3729 (Lords, Oct 26): Amendment 41 ‘Reporting of incidents involving frontier AI models’, moved by Baroness Harding of Winscombe (Con) with Lord Clement-Jones, Baroness Kidron and Baroness Berger: Contents 95, Not-contents 165; government win. Contents: 49 Liberal Democrat, 26 crossbench, 10 Conservative; Not-contents: 158 Labour (Lords votes record)
- Amendment 41 would have required a developer that is, or ‘ought reasonably to be’, aware that its frontier model gained or was used to gain unauthorised access to a regulated service to notify the service and regulator within 24 hours, update within 72 hours and report within 14 days, saying whether the model acted autonomously (marshalled list; Harding and Tarassenko in debate)
- Government Amendment 70 (Baroness Lloyd of Effra): regulations may confer functions on the Secretary of State where a compromise ‘is or may be attributable to loss of oversight or control of an AI agent’: investigating, obtaining and preserving information, testing remedies; ‘AI agent’ and ‘loss of oversight or control’ to be defined by regulations (affirmative procedure, Amendment 87)
- Clement-Jones welcomed Amendments 70 and 74 as ‘a first acknowledgement by the Government that it is not enough to be technology-agnostic’, but called them ‘entirely reactive and diagnostic: gathering information only after the horse has bolted’
- Minister Baroness Lloyd: ‘A broad power to shut down AI systems could have implications for investment, growth and our digital sovereignty’; existing NIS direction powers could require isolating systems, restricting data flows or suspending an AI system’s automated actions; a consultation on incident reporting will follow
- Other AI amendments debated in the same group: Clement-Jones’s Amendment 65 (‘last-resort’ shutdown powers over data centres and large AI deployments), Tarassenko’s Amendment 66 (AISI pre-deployment audits of frontier AI used in critical infrastructure), Kidron’s Amendment 67 (red lines on autonomous cyber attacks, evading shutdown, self-replication); no division on them is recorded for Oct 26
- Harding cited OpenAI’s access to Australian Medicare data in June, reported to Australia only in September, and said OpenAI and Anthropic representatives (including Lord Booth-Smith for Anthropic) had backed required reporting ‘at least in principle’ before a Commons committee the week before
- Tarassenko: AISI contained its own summer unsanctioned-agent incident within an hour, ‘confirmed to us by Henry de Zoete at a meeting in Portcullis House today’
Show 2 more
- The same evening the government lost two non-AI divisions on the bill (Lord Birt’s NIS-effectiveness review, 202–155; Viscount Camrose’s register of foreign powers, 194–155). Third reading: Nov 2, 2026 (Parliament bills API)
- Nov 2: Lords third reading held (Parliament bills API); the bill returns to the Commons for the Lords amendments
What happened
The Cyber Security and Resilience Bill had left committee with AI outside its scope. For report stage the government tabled Amendments 70 and 74, its first AI-specific clauses: powers, to be set out in regulations, to investigate and remedy security compromises attributable to the “loss of oversight or control of an AI agent”. Peers from four parties argued that this put duties on the victims (regulated essential services) but none on the frontier developers whose agents cause the incidents. Baroness Harding pressed Amendment 41 to a vote and lost 95–165, with Labour peers voting against almost en bloc.
The minister, Baroness Lloyd of Effra, set out the government’s line: AI risks need global rules (the UK’s G20 presidency next year), commercial AI services delivered as cloud or software services are likely already in scope of the NIS regime, and existing national security direction powers can order isolation or suspension in a loss-of-control scenario, which she called more proportionate than a broad shutdown power.
The record was read from Parliament’s own data (Lords votes, Hansard and bills APIs). No press report of the vote had appeared by the morning of Oct 27.
Why it matters
It is the first time the UK has put AI-agent loss of control into statute, but it is reactive: the obligation to report sits with the attacked service, not the lab. A legal duty on labs to report rogue-agent incidents, already law in New York (RAISE Act) and California, was rejected for now.
Sources
5 sources from 4 sites. Numbers match the chips in the text.
5 sources: 5 primary
Primary
- UK Parliament Lords votes: division 3729, Amendment 41 (frontier AI incident reporting), 26 Oct 2026votes.parliament.uk, official
- Hansard (Lords, 26 Oct 2026): Cyber Security and Resilience (Network and Information Systems) Bill, reporthansard.parliament.uk, official
- HL Bill 49-I: marshalled list of amendments for reportbills.parliament.uk, official
- UK Parliament: bill stages (report Oct 26, third reading Nov 2)bills.parliament.uk, official
- UK Parliament bills API: stage record (Nov 3)bills-api.parliament.uk, official
Changes
- Third reading on Nov 2 noted (due item)
- Filed from the Lords division record, Hansard and the marshalled list