UK AI safety researchers are slated to receive computing worth $250 million under NVIDIA's five-year pledge, equivalent to about $50 million a year on average; the support is capacity rather than grant payments.
The plan calls for UK cloud partners to provide the computing. Their names and the terms of their arrangements have yet to be settled.
BOLD, led by Oxford University, and SOFAIR, led by University College London (UCL), are part of the work, as is Imperial College. Researchers will study AI agent safety, open-model security and ways to defend against threats in real-world settings.
Five British startups are named for planned investment: Boltz, Foresight, Hiverge, Prima Mente and Riverlane. Boltz develops AI models for biomolecular applications, while Foresight focuses on forecasting infrastructure risks. Hiverge is creating an AI tool to optimize algorithms.
Prima Mente is working on biological foundation models aimed at neurodegenerative diseases. Riverlane's work combines hardware and software for quantum error correction.
