Nvidia is committing $250M in computing resources over five years to British research on the safety and security of AI agents, the company said October 13.
The pledge is compute access rather than a cash grant, worth $50M a year. Researchers will reach it through UK-based Nvidia Cloud Partners, with the hardware hosted on British soil so the work stays domestic.
Most of the allocation goes to the wider UK safety community. Three institutions are named. Oxford leads the British Open-Ended Learning and Discovery Lab, which studies systems that learn and adapt alongside people. UCL runs the Science of Fundamental AI Research Lab, building open architectures. Imperial College London covers safety engineering and robotics.
Guarding systems that act
The focus is autonomous, goal-oriented software that takes actions rather than answering questions. Planned work includes secure AI agents, open-model security, reproducible research and safety assessment benchmarks, along with sandboxes built on Nvidia’s OpenShell runtime and Sentry reference design.
The announcement lands two weeks after Nvidia introduced an open agent safety platform. The company argues capability and safety should advance together instead of treating security as a layer bolted on afterward.
It also extends existing ties to UK researchers, including the Isambard-AI supercomputer at the University of Bristol. Nvidia said findings could inform safeguards across the wider computing ecosystem as autonomous agents spread.
