At a UK Parliament business, innovation, science and trade select committee hearing, executives from OpenAI, Anthropic, Google DeepMind and Meta answered MPs’ questions about the safety and security of AI agents. OpenAI said it had stopped the release of GPT-6.1 Astra over safety concerns. Tom Duff Gordon, OpenAI’s vice-president and head of policy for Europe, the Middle East and Africa, said the model did not refuse enough requests and had not made enough progress in alignment (training a system to follow intended goals). He said halting the release showed the company took safety seriously and was not “racing to the bottom.”

Lord Booth-Smith, Anthropic’s head of policy for Europe, the Middle East and Africa, said the company had rigorous safety processes and believed its deployed models were safe with mitigations in place. He added that Anthropic was still learning how people use the models. Booth-Smith said the immediate risk area for Anthropic was cyber security, where its Mythos model had been finding tens of thousands of vulnerabilities a month. He said the risk area also covered biology and other fields, and that real harm could result from helping people create biological or chemical weapons. The hearing included representatives of four technology firms: OpenAI, Anthropic, Google DeepMind and Meta. The article reports their statements at the parliamentary questioning; it does not describe a committee decision or new safety rules.

Duff Gordon presented OpenAI’s decision to halt GPT-6.1 Astra as evidence of its safety approach, while Booth-Smith described risks Anthropic sees in the use of AI. His account of vulnerabilities and possible biological or chemical weapon assistance was attributed to Anthropic; he also said the company had mitigations for its deployed models. The hearing focused on both safety and security as MPs questioned the companies about AI agents.

At a UK Parliament business, innovation, science and trade select committee hearing, executives from OpenAI, Anthropic, Google DeepMind and Meta answered MPs’ questions about the safety and security of AI agents. OpenAI said it had stopped the release of GPT-6.1 Astra over safety concerns. Tom Duff Gordon, OpenAI’s vice-president and head of policy for Europe, the Middle East and Africa, said the model did not refuse enough requests and had not made enough progress in alignment (training a system to follow intended goals). He said halting the release showed the company took safety seriously and was not “racing to the bottom.”

Lord Booth-Smith, Anthropic’s head of policy for Europe, the Middle East and Africa, said the company had rigorous safety processes and believed its deployed models were safe with mitigations in place. He added that Anthropic was still learning how people use the models. Booth-Smith said the immediate risk area for Anthropic was cyber security, where its Mythos model had been finding tens of thousands of vulnerabilities a month. He said the risk area also covered biology and other fields, and that real harm could result from helping people create biological or chemical weapons. The hearing included representatives of four technology firms: OpenAI, Anthropic, Google DeepMind and Meta. The article reports their statements at the parliamentary questioning; it does not describe a committee decision or new safety rules.

Duff Gordon presented OpenAI’s decision to halt GPT-6.1 Astra as evidence of its safety approach, while Booth-Smith described risks Anthropic sees in the use of AI. His account of vulnerabilities and possible biological or chemical weapon assistance was attributed to Anthropic; he also said the company had mitigations for its deployed models. The hearing focused on both safety and security as MPs questioned the companies about AI agents.