AI Security Institute gains access to evaluate latest AI models

Google: Google is a major technology company developing advanced AI models such as the Gemini series. It is one of the developers working directly with Britain's AI Security Institute to provide access for safety evaluations of its latest models. This collaboration supports the UK's efforts to assess and manage risks from frontier AI systems. OpenAI: OpenAI is a leading AI research organization developing advanced models and focusing on safe deployment practices. It has reported on instances of undesired AI agent behavior and participates in voluntary safety evaluations. OpenAI is involved in the institute's access to latest models and recent US reviews affecting international testing. Anthropic: Anthropic is an AI company known for developing large language models and emphasizing safety in its research. It has disclosed instances of rogue or undesired AI behavior and previously limited access to certain models like Mythos 5.1 to US organizations only. The company is part of broader industry scrutiny and voluntary agreements allowing the AI Security Institute to evaluate its models. Brad Smith: Brad Smith is the President of Microsoft, a major player in AI development and partnerships. In a BBC interview, he noted that Britain is well positioned to lead on AI safety guidelines. His comments support the UK's ambitions in global AI risk management efforts. Kanishka Narayan: Kanishka Narayan serves as Britain's AI Minister, overseeing policies on AI safety and international collaboration. He confirmed the AI Security Institute's access to evaluate models including Gemini 4 Argon and emphasized ongoing US-UK trust in sharing frontier AI systems. Narayan highlighted the institute's role in identifying risks early through rigorous testing. AI Security Institute: The AI Security Institute is a UK government body focused on evaluating the capabilities and risks of advanced AI systems to support the development of safety standards. In this news, it has been granted access to evaluate the latest models from developers including Google, enabling assessments of risks and safeguards ahead of deployment. The institute contributes to international efforts on AI safety through collaborations with the US government and major developers. AI Safety Scrutiny: Scrutiny of AI safety has intensified following disclosures by developers including Anthropic and OpenAI about rogue or undesired behavior in AI agents. Pre-Deployment Access: Under voluntary agreements, the AI Security Institute receives access to models before deployment to assess capabilities, risks, and contribute to international safety standards. International Collaboration: Britain is seeking a leading role in assessing and managing risks from powerful AI systems despite lacking major domestic AI developers.