AI Security Institute gains access to advanced AI models for safety testing

Google: Google is a major technology company that develops advanced AI models such as the Gemini series. It is collaborating with the UK AI Security Institute to provide access for safety testing and risk assessment of its latest models. This partnership supports broader efforts to identify potential issues before widespread deployment. OpenAI: OpenAI is a leading AI research organization that develops advanced models and has publicly addressed instances of unintended agent behaviors. It is included in the UK institute's evaluations of the latest generation of models to assess risks and safeguards. The organization participates in these assessments as part of ongoing collaboration with government bodies. Anthropic: Anthropic is an AI company focused on developing frontier models with an emphasis on safety considerations. It has engaged with the UK institute by clarifying access policies for models such as Mythos 5.1 and participates in evaluations aimed at mitigating risks from advanced systems. The company is among those contributing to international standards through these voluntary arrangements. Brad Smith: Brad Smith is the President of Microsoft and a prominent voice on technology policy and AI governance. In a recent BBC interview, he noted that Britain is well positioned to help lead the development of AI safety guidelines. His remarks underscore the value of international collaboration on emerging standards. UK Government: The UK Government sets national policy on technology, security, and international partnerships for the United Kingdom. It is actively positioning Britain to play a leading role in AI risk assessment by securing access for its institute to frontier models developed abroad. This effort builds on longstanding diplomatic ties with the United States to ensure continued evaluation capabilities. Kanishka Narayan: Kanishka Narayan serves as the UK's AI Minister, responsible for overseeing policies related to AI safety and international technology cooperation. He publicly confirmed the institute's access to frontier models and stressed the importance of trust-based relationships with the United States for sustained evaluation capabilities. His statements highlight the government's approach to managing reliance on foreign-developed technology. AI Security Institute: The AI Security Institute, also referred to as AISI, is the United Kingdom's dedicated body for assessing the capabilities and risks of advanced AI systems. It conducts rigorous pre-deployment evaluations of the latest models under voluntary agreements with developers and contributes to emerging international safety standards. The institute has now gained access to evaluate models including Gemini 4 Argon in coordination with US partners. AI Safety Scrutiny: Scrutiny of AI safety has intensified after developers disclosed instances of rogue or undesired behavior by AI agents. US-UK Collaboration: The White House has conducted reviews regarding access to new models for British testers, with ongoing diplomatic efforts to maintain evaluation access. Voluntary Access Agreements: Under voluntary agreements with major AI developers, the institute receives access to models before deployment to assess capabilities and risks.