Asked by: Lord Nagaraju (Labour - Life peer)
Question to the Cabinet Office:
To ask His Majesty's Government, further to the Written Statement by Baroness Twycross on 7 September (HLWS321) in which they stated that the risks from increasingly powerful artificial intelligence systems are “inherently transnational” and that no country can address them alone, what steps they are taking with international partners to develop shared arrangements for (1) the evaluation of advanced AI systems, (2) notification and information-sharing concerning serious AI incidents, and (3) coordinated responses to risks arising from advanced AI systems that cannot be reliably controlled or directed; and what role they envisage for the United Nations Global Dialogue on Artificial Intelligence Governance in developing such arrangements.
Answered by Baroness Twycross - Parliamentary Secretary (Cabinet Office)
AI is already subject to UK law and regulation through existing laws and expert regulators. But the most advanced AI systems present risks that do not stop at national borders, and we know that these challenges cannot be addressed by a single country alone. That is why the Prime Minister has made artificial intelligence a central focus of Britain’s G20 Presidency next year, and committed to work towards principles and standards that ensure the most capable AI systems are developed securely and with appropriate transparency. The AI Security Institute (AISI) will play a key role in delivering this global commitment, with partners across industry, governments and beyond.
Asked by: Lord Nagaraju (Labour - Life peer)
Question to the Cabinet Office:
To ask His Majesty's Government what assessment they have made of Anthropic’s decision not to provide its Claude Mythos 5.1 model to the UK AI Security Institute for pre-release testing; whether the Institute has subsequently been provided with access to that model; and what steps they are taking to ensure that the Institute has timely access to frontier AI models for independent safety evaluation.
Answered by Baroness Twycross - Parliamentary Secretary (Cabinet Office)
AISI maintains strong relationships with all frontier AI developers, that whilst voluntary, are bound by memoranda of understanding, and continues to test the world’s most capable models prior to their deployment. Only this month, we tested OpenAI’s most powerful model, GPT-6 Astra, before it was released to the public. AISI has tested dozens of models from leading AI companies, including OpenAI, Google DeepMind and Anthropic.