Jensen Huang Draws the AI Safety Red Line: If Customers Develop Unsafe Software, Nvidia May Cut Off Chip Supply

Beating AI News: NVIDIA CEO Jensen Huang recently said in an interview with Bloomberg in London that, under a hypothetical scenario, if a customer develops unsafe AI software, NVIDIA could stop supplying chips to them; however, before taking action, it would first communicate with the customer. Huang also emphasized that NVIDIA provides general-purpose technology, and the company has certain limitations in judging how customers use this technology.

Meanwhile, NVIDIA has committed to providing $250 million worth of computing resources to the UK over the next five years to support AI safety research, with a focus on agent safety testing, with the related resources valued at about $50 million per year. Supported projects include the BOLD project led by the University of Oxford, as well as related research at Imperial College London and University College London.

Huang advocates accelerating the integration of safety constraints and monitoring technologies into AI systems, and stated that AI capabilities and safety must advance together. The report noted that AI safety regulation could affect NVIDIA's chip demand: if regulation restricts new model training or drives up development costs, it could dampen purchases of AI processors; however, strengthened safety research could also help promote more reliable deployment of AI technology.