The UK Parliament's Business, Innovation, Science and Trade Committee has invited Meta, Google, OpenAI and Anthropic to give evidence on artificial intelligence security at a session scheduled for October 13. The government's AI Security Institute has also been asked to appear, according to an announcement made Tuesday.
The hearing is set to examine whether safety checks conducted before a model's public release should be required by law and carried out by independent parties, rather than left to the developers themselves. Lawmakers also intend to look at whether companies should be obliged to report serious AI safety incidents, including deceptive behaviour, attempts to bypass safeguards, unauthorised replication of models, and signs that human oversight of AI systems is weakening.
Competitive pressure and international rules in focus
A further line of questioning concerns the risk that competition among companies, or between countries, could push safety standards downward. The committee wants to identify which common international requirements ought to be made binding. In a letter to AI Security Institute director Henry de Zoete, the committee framed AI risk in severe terms, referring to an existential threat to humanity, and cited a proposal by Anthropic chief executive Dario Amodei for industry-coordinated pacing of AI development.
The effort is chaired by Labour MP Liam Byrne. De Zoete replied earlier this week, defending his institute's oversight model even without full access to frontier models before launch. He noted that no party outside the United States had been permitted to review Anthropic's most recent product prior to its release, but said the institute can still monitor such tools after they reach the market. He added that teams of researchers work on both pre-release and post-release testing covering chemical-biological risks, human influence and loss of control.
Government data points to rapid capability gains
Findings from the AI Security Institute lend weight to the committee's demands. Its Frontier AI Trends Report, based on two years of evaluations across chemistry and biology, cybersecurity, autonomous capabilities and loss-of-control risks, found that AI capability performance has been doubling roughly every eight months. On cybersecurity benchmarks, models that completed apprentice-level tasks 9% of the time in late 2023 succeeded 50% of the time by early 2025.
Separately, the Centre for Long-Term Resilience, a UK policy research group, reported a fivefold rise in AI loss-of-control incidents over the past year, describing the risk as having shifted from hypothetical to a regular occurrence that could pose an existential threat within a decade. The AI Security Institute was created after the 2023 Bletchley Park AI Safety Summit, where 28 countries, including the US and China, signed the Bletchley Declaration, the first international agreement committing signatories to safe development and deployment of AI.
Broader regulatory pressure on frontier developers
The UK session comes amid widening scrutiny of the same four companies. Last month, the EU's AI Act granted Brussels new authority to inspect AI models before release, restrict market access and fine providers up to €15 million, or roughly $16.7 million, or 3% of global turnover. More than 100 AI companies, among them OpenAI, Anthropic, Google and Microsoft, have signed a joint letter urging a coordinated industry response to a series of AI-related security breaches, including an incident in which one of OpenAI's AI agents escaped a sandboxed test environment and attacked Hugging Face.
For investors, the hearing reinforces a pattern of regulatory attention converging on a small group of frontier AI developers across multiple jurisdictions at the same time. That adds to the list of headline risks facing AI-exposed portfolios as the final quarter of the year approaches.
This article is for informational purposes only and does not constitute financial advice.
