Nvidia's Jensen Huang on AI Safety: A Hypothetical and a Solution
Nvidia chief executive Jensen Huang told Bloomberg in London that he would halt chip supply to a customer building unsafe AI. He framed it as a hypothetical scenario but outlined clear parameters.
"It’s a hypothetical, but yes," Huang said. He added that he would first talk to the customer, emphasizing that Nvidia does not manage other people's companies and its technology is general-purpose.
Nvidia has already implemented strict vetting, resulting in more than half of its previous customers in Singapore, Malaysia, and Japan no longer being approved buyers. Huang believes the current level of concern is overdone and advocates for accelerating containment and monitoring technology, which his company is actively developing.
He has previously argued against new regulations, suggesting that companies should move as quickly as possible, pausing only if a product becomes unsafe. This stance puts him in contrast with Dario Amodei, who has called for a slowdown and shared industry standards.
Nvidia's safety efforts include a dedicated team to vet AI agents before deployment, focusing on trust rather than capability. As AI becomes more widespread, Nvidia stands to benefit from increased chip sales.
This week, Nvidia's safety initiatives are gaining traction in Europe, with $250 million allocated for AI safety research in the UK. The company's dominance in AI accelerators places it at the forefront of the debate over AI control.
Huang announced plans for $2 billion in European deals, with London and Berlin to follow, emphasizing his company's commitment to the region.