In late September 2026, the U.K. Parliament’s Business, Innovation, Science and Trade Committee formally invited OpenAI, Anthropic, Google DeepMind and Meta to an AI-security evidence session on 13 October. Coverage from City A.M., MLex and regional press confirms MPs want to question the labs on mandatory independent pre-release testing, AI Security Institute (AISI) access to models, serious-incident reporting, and whether regulators should be able to delay or block deployments.
Companies must confirm attendance by 29 September. AISI is also invited. The timing is deliberate: it lands weeks after public calls to slow frontier development, reports of a possible OpenAI–Anthropic stress-test pact, and a wave of EU AI Office RFIs to more than 30 providers.
British lawmakers face a classic tension. London wants to attract lab and data-center investment, yet faces political pressure after a summer of advanced-model incidents and researcher warnings about loss of control. A televised hearing with the sector’s four most visible names turns that tension into democratic theater—and a possible prelude to harder legislation.
The announced questions go to the operational core: who tests before release? Does AISI get real access or only summaries? What threshold triggers an incident report? Can the state impose a deployment pause when safeguards lag capabilities? Those four questions are the same ones Amodei and Altman discuss in essays and interviews, now under parliamentary oath.
For Meta and Google DeepMind, the summons also rebalances media focus that often concentrates on OpenAI and Anthropic. Meta brings the open-weight and consumer-agent dimension; DeepMind brings science and Gemini models. The committee appears to want a sector map, not a bilateral showdown.
AISI’s invitation is an institutional signal. Britain’s AI security institute has been cast as sovereign evaluation capacity; the hearing will test whether labs accept its technical access or dilute it with NDAs and controlled demos. Without access to weights, fine-tuning tools and incident logs, AISI remains a paper auditor.
Politically, London navigates between competing with the EU for lab headquarters and not looking lax. The EU AI Act already applies RFI powers and fines; the U.K., outside that regime, needs to show teeth of its own. A 13 October session with 29 September confirmations is maximum-pressure scheduling for lab counsel.
U.K. enterprise buyers should pay attention: whatever emerges—reporting commitments, pause thresholds, AISI access—may foreshadow contractual duties in public procurement and regulated sectors. Do not wait for final statute to inventory agents and incident channels.
The summons also interacts with the U.S. antitrust suit over an alleged slowdown agreement: if in London labs defend more testing and possible pauses, in Washington they may be accused of coordinating to restrain trade. The rhetorical craft will be distinguishing safety cooperation from competition restriction.
Neural Pulse sees the hearing as the first major parliamentary confrontation after the summer of agentic incidents. Labs have spoken to legislators before, but rarely with an explicit agenda of deployment-blocking power.
Session risks include grandstanding without legislative follow-through; vague trade-secret answers; or, conversely, public commitments that later collide with Q4 2026 commercial roadmaps. Value will sit in written evidence the committee publishes afterward, not only in a 30-second clip.
In short, the U.K. Parliament set 13 October 2026 as the appearance date for OpenAI, Anthropic, Google DeepMind and Meta, with AISI in the room and confirmations due by 29 September. The agenda: testing, access, incidents and pauses.
We will track confirmation letters, the final witness list and the committee’s pre-briefs. If a lab declines, that will be news too.
For the British public, the session will be the thermometer of whether “frontier safety” means voluntary papers or real state power over the deploy button.
