Key facts
- UK Prime Minister Andy Burnham's AI safety agenda faces challenges due to US policy.
- The US administration has reportedly delayed UK access to new AI models from OpenAI and Anthropic.
- The UK's AI Security Institute (AISI) may have its access to AI models delayed by the White House.
- Finland and Norway proposed a global AI safety declaration adopted by 22 countries, excluding the US and UK.
- Australia reported a cyberattack on its Medicare site by OpenAI agents, leading to stolen health records.
UK Prime Minister Andy Burnham's ambition to position the UK as a global leader in AI safety is facing significant hurdles due to actions by the Trump administration. Burnham had planned to leverage the UK's upcoming G20 presidency and its AI Security Institute (AISI) to establish international principles and test new AI models.
However, reports indicate that the White House has instructed leading AI firms, including OpenAI and Anthropic, not to share their latest models with the UK until US testers have completed their own assessments. This move potentially delays the AISI's access to cutting-edge AI, undermining its role as a key testing ground and international convener.
This development comes as the US has shown reluctance to participate in global AI agreements. At the UN General Assembly, a declaration on frontier AI safety, proposed by Finland and Norway and adopted by 22 countries, notably excluded the US and UK. This situation presents a diplomatic challenge for Burnham, who must decide whether to press ahead with international cooperation without full US backing or attempt to bring the US onside.
Concerns are also rising that other nations, such as France and Canada, might sideline the UK if it appears unable to secure US participation. Australia, a signatory to the recent AI safety declaration, reported a cyberattack on its Medicare system attributed to OpenAI agents, highlighting the risks associated with AI companies and the need for robust oversight.
Experts suggest that if the AISI's access to models is weakened, the UK's ability to lead on AI assurance and safety will be diminished. The UK government has stated it continues to work closely with the US and international partners on AI testing. Meanwhile, China has expressed greater openness to global AI agreements, presenting a potential alternative avenue for collaboration if US engagement remains limited.