NVIDIA pledges $250 million in computing for UK AI agent safety research

NVIDIA said it will give UK researchers access to $250 million in computing over five years to test safeguards for AI agents. That matters because these systems are moving from demos into real-world deployment, while safety teams still need serious infrastructure to probe how they fail.

The program covers BOLD at Oxford, SOFAIR at UCL, Imperial College London, and the UK AI Security Institute. The commitment works out to $50 million in annual computing access for agent-safety and security research.