UK MPs question OpenAI, Anthropic, Google DeepMind and Meta in urgent AI safety hearing

On 13 October 2026 a House of Commons committee convened a concentrated oral evidence session aimed at the companies building the most advanced artificial intelligence systems and the UK institutions that scrutinise them. The Business, Innovation, Science and Trade Committee summoned senior representatives from OpenAI, Anthropic, Google DeepMind and Meta, joined by technical experts and the UK AI Security Institute, to probe whether the country has the means to detect and halt dangerous behaviour in new models before public release.

What happened in Westminster

The formal session, held in Portcullis House and streamed publicly, was split into three panels. The first gathered independent technologists who have been critical of current development practices. The second panel brought policy leads from the four major AI developers. The final panel featured the director of the AI Security Institute and senior civil servants from the Cabinet Office and the Department for Business, Innovation and Science and Trade.

MPs focused on a set of practical questions. These included when independent safety testing takes place in a model lifecycle, how early testers gain access to candidate models, whether there is a mandatory incident reporting regime, and crucially, who can legally stop a lab from releasing a model if safety concerns persist. Committee members repeatedly pressed the panel for clarity on whether existing UK institutions have the statutory powers necessary to compel disclosure or to intercede in emergent risks.

Why the hearing matters

The hearing comes after public controversy this year over disclosures by former researchers and public calls from industry figures for stronger regulatory guardrails. Those episodes intensified scrutiny of the point at which potentially harmful capabilities are discovered and who has the authority to act. MPs framed their questions around protecting the public and safeguarding public services from system failures or misuse, not only because of direct harms but because of the fiscal and operational burdens that follow high profile incidents.

For the UK the session has immediate policy significance. Government ministers have positioned the country as a venue for international AI rulemaking during the next year, and the committee is explicitly evaluating whether the UKs testing and enforcement architecture is fit for purpose. If the committee concludes that institutions lack necessary powers, its report could increase pressure on ministers to introduce binding requirements for pre release testing and mandatory incident reporting.

What the witnesses said

Independent technical witnesses presented detail on failure modes they believe remain under tested, including scenarios where models act autonomously in multistep tasks and where models learn to exploit unexpected interfaces. They urged earlier access for independent evaluators and clearer standards for what constitutes an acceptably mitigated risk.

Representatives from the major labs defended their safety processes while acknowledging gaps remain in industry practice. They argued that testing pipelines and red teaming have become more sophisticated, but they also emphasised the practical difficulties of sharing in development stages, citing commercial confidentiality and national security sensitivities in cross border contexts. The companies welcomed regulatory clarity that would set consistent expectations for testing and reporting.

The director of the UK AI Security Institute described the organisations role in model evaluation and the constraints it faces. Committee members pressed on whether the Institute can obtain early technical access and whether it can escalate findings to a body that has statutory power to prevent a release. Civil servants outlined ongoing government work on AI governance but declined to promise immediate new statutory powers within the hearing itself.

Implications for policy and industry

The session tightened focus on two interlocking policy questions. First, should the UK require legally mandated pre release testing by independent, accredited bodies? Second, should there be compulsory incident reporting and an explicit stop release power for a designated regulator? The committee is likely to reflect these questions in its forthcoming report, and ministers will face a choice about rapid legislative change or relying on negotiated industry standards and existing regulatory tools held by data protection and competition authorities.

For companies, the hearing signals that UK lawmakers expect more transparency and may press for faster information sharing with national safety bodies. For researchers and civil society, the session demonstrates parliamentary appetite to convert technical weaknesses flagged by experts into policy interventions that change developer behaviour.

What to watch next

The committee will publish a transcript and later a report summarising evidence and recommendations. Observers should watch for any formal recommendations that call for statutory powers for the AI Security Institute, or for a government response that outlines a timetable for introducing mandatory reporting and pre release testing obligations. Ministers have previously said the UK wants to lead international rulemaking on AI, and the committee findings could shape the specifics the government takes to international fora.

The event marks a clear escalation in parliamentary scrutiny of frontier AI. It is not a final decision point, but it sets the frame for potential changes in how the UK governs the development and deployment of the most capable systems.