UK to Test OpenAI and Anthropic Models Again Before Release
The UK government has stated that it will continue to test the most powerful AI models, including those from OpenAI and Anthropic, before they are released. AI Minister Kanishka Narayan made this announcement on Monday, reversing a recent US decision to prioritize White House checks.
Narayan emphasized his commitment to making the UK one of the hardest countries to attack using artificial intelligence. He attributed this change to close collaboration with the US and the recent King Charles summit, which brought together ministers and tech leaders.
However, Narayan remained cautious, noting that access to cutting-edge AI technology may not be permanent.
Recent Developments
Last month, the White House requested that OpenAI and Anthropic hold back new model releases to UK testers until US officials had conducted their own checks. Anthropic opted not to share its Mythos model with the UK's AI Security Institute, according to Bloomberg.
The Telegraph reported on the UK's successful lobbying efforts to reverse this decision.
AI Security Task Force
In addition to testing, Narayan also established a task force to safeguard UK government systems from AI-driven attacks. He revealed that two pilots using frontier AI identified over 1,000 potential vulnerabilities.
The US has its own pre-release review process, with tech giants like Google, Microsoft, and xAI agreeing to conduct these checks.