A forthcoming white paper will outline government and infrastructure operators’ requirements from advanced AI systems and clarify responsibilities for AI model providers.

Singapore’s IMDA and Microsoft have signed a memorandum of understanding to collaboratively advance AI safety, security, and trustworthy deployment, with a focus on testing frontier AI models for safety across languages.

The partnership, announced in Singapore, is framed as a government regulator–tech company cooperation to raise AI safety standards and governance.

The joint effort aims to foster evaluation sciences, develop benchmarks, and establish concrete evaluation methods to support responsible AI governance.

Industry context includes Anthropic’s Claude Mythos Preview revealing numerous vulnerabilities and limited public access via Project Glasswing, drawing regulatory attention from major institutions.

The collaboration seeks to balance AI innovation with safety and reliability, addressing risks like cyberattacks and misinformation as frontier AI becomes more accessible.

Microsoft Chief Responsible AI Officer emphasizes leveraging government insights alongside technical expertise to strengthen AI evaluation and address emerging risks.

Early access provisions to models are seen as a way to remediate security flaws and help trusted partners fortify defenses against AI-enabled exploits.

Key voices include IMDA’s Kiren Kumar stressing public-good governance, and Microsoft’s Natasha Crampton noting the collaboration’s role in shaping global AI safety conversations.

The agreement prioritizes technical AI safety research, including agentic AI, safety evaluation benchmarks, multilingual safety, and public resilience against misinformation.

Kiren Kumar emphasizes building governance, evaluation benchmarks, and practical tools beyond policy to support responsible deployment.

The partnership will include knowledge sharing of governance frameworks and results through the Singapore AI Safety Institute and other agencies to explore responsible use of frontier AI by governments and critical infrastructure operators.