Nvidia and Meta bosses reject efforts to co-ordinate AI slowdown

Nvidia’s Jensen Huang and Meta’s Mark Zuckerberg have pushed back against efforts by “frontier” AI labs to co-ordinate a slowdown in research in the name of safety, arguing companies should address the risks individually.

The two chief executives in separate comments on Tuesday emphasised the responsibility of individual companies to tackle concerns that AI research risks spinning out of control and posing a threat to humanity.

Huang, who has been dismissive of such existential warnings, went further, saying there was no need for new US laws or regulations on AI safety, and urging AI labs not to fall into a “false choice” between the speed of innovation and the ability to develop safe products.

The pair of tech founders weighed in after Anthropic’s Dario Amodei, OpenAI’s Sam Altman and Elon Musk on the weekend publicly called for a co-ordinated slowdown in research to properly tackle the risks AI poses.

The rare moment of unity among the three bitter rivals followed mounting public outcry and warnings from researchers that an all-out race to advance AI risked humans losing control of the technology with catastrophic consequences.

Zuckerberg, who has created a major AI lab inside Meta, and Huang, who leads the world’s biggest supplier of AI hardware which offers its own suite of AI models, have not signed up to those proposals for industry co-operation to “pace the frontier”, even as Google and Microsoft have signalled support.

Their split reflects deeper divisions in the AI industry between the leading developers of proprietary “closed” models, such as OpenAI and Anthropic, and those advocating for customisable “open” models.

Huang has pushed the importance of open models to encourage widespread AI adoption, while Meta has released some closed models but generally argued for open systems. Some open model advocates say regulations or controls on development would advantage big labs that sell closed systems.

Huang on Tuesday argued companies should individually focus on not releasing dangerous products.

“The market forces are already there, we don’t need any new laws, we don’t need new regulations,” he said at an event with Salesforce’s chief executive Marc Benioff in San Francisco on Tuesday.

“If you build a product or a service and you’re not confident in its functionality, capability or safety, then don’t release it,” said the Nvidia boss, whose company is a big investor in OpenAI, Anthropic and Musk’s SpaceX. “That’s a very obvious thing to do: You pace yourself until you are confident you’re releasing something that the market will appreciate.”

“You could definitely have both [innovation and safety] at the same time,” Huang added. “So run as fast as you can, but if you feel at any given point in time the company is out of control or products are not going to be safe, you know, take a pause and get it right.”

The head of the world’s most valuable company has aligned himself with Donald Trump against an AI slowdown. Huang took a phone call from the president while on stage at an event on Monday, where Trump again rejected existential concerns about the technology as a “hoax” and railed against moves to slow US research and data centre construction.

Recommended

“You’re right. We’re not going to let that happen, sir,” Huang responded.

Zuckerberg outlined his views in a post on X later on Tuesday, echoing Huang’s argument that the responsibility to ensure safety lies with individual AI labs.

“Every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens,” Zuckerberg wrote. “Labs face significant liability if their models cause harm, so they have a strong incentive to prevent this as well.”

Zuckerberg said Meta delayed shipping its own Muse AI agent for several months but “didn’t call for everyone else to do this before we would”.

He added that a “larger and more diverse ecosystem of evaluators”, the independent experts used by labs to assess models, would be helpful. But he did not address proposals from Anthropic to give these evaluators greater access.