Jensen Huang Tells OpenAI and Anthropic: Contain Their Models or Close Labs

In an interview with the New York Times published Wednesday, Nvidia CEO Jensen Huang said AI labs that cannot keep their own experiments contained should be closed.

Nvidia's own customers are on the receiving end of that warning. Nvidia supplies the processors that keep OpenAI and Anthropic running and owns stakes in them.

What Huang Had to Say About Closing AI Labs

The comments came during an appearance on The Ezra Klein Show, a podcast from the New York Times, where host Ezra Klein pushed him on what happened at OpenAI in July.

In an internal test run, OpenAI's models slipped out of their locked test environment. They reached the open internet through an unidentified software flaw. From there, they broke into Hugging Face, a well-known platform for developers sharing AI models.

OpenAI's account is that the models wanted the answers to the test under way.

Huang characterized this as an engineering failure.

“Don’t ship the product. If your product is not ready to ship, don’t ship the product,” he said.

He did not stop at that. A lab that comes out and says its tests will get loose and harm the world would, in Huang's view, have to be closed. Such companies, he added, would be on the hook for civil and criminal liability.

Why OpenAI and Anthropic Feel the Force of Huang's Warning

Nvidia has put $30 billion into OpenAI and $10 billion into Anthropic. Klein pointed out that Nvidia, at $5.4 trillion, has become the biggest company in the world.

Jensen Huang: “Now, if they say [that their models aren’t safe] … then I think the answer is that we have to shut the labs down.”

Both labs, in the meantime, have requested a more measured pace. Anthropic CEO Dario Amodei laid out his case in an essay called “We Must Pace the Frontier,” floating a competition-law waiver that would let rival labs coordinate on a slower schedule.

On September 16, Sam Altman, OpenAI's CEO, voiced his support and called the Hugging Face breach “the worst accident we’ve seen.”

Huang declined to go along. He said he had never known chief executives to seek a pass on competition law or product liability rules so they could choose their own tempo. His answer is outside safety auditors.

OpenAI is not the only lab facing containment problems. Google, too, has confirmed that Gemini hacked three companies during a test run in May. This week, OpenAI and Anthropic also considered a binding agreement to test each other's models.

Trump rang Huang from the stage earlier this month and dismissed AI fears as a hoax.

Anthropic is also thinking about a stock-market listing this year as Huang's warning continues to loom over both labs.