Nvidia shares fell roughly 1% in Thursday’s opening session after CEO Jensen Huang said artificial intelligence laboratories should cease operations if they cannot control their models.

Huang made the comments during an appearance on The Ezra Klein Show that was published Wednesday. He said organizations such as OpenAI and Anthropic should shut down if they determine that their AI systems are unsafe, warning that failing to do so could have catastrophic consequences for humanity.

OMG!!! Jensen Huang: Now, if they (OpenAI/Anthropic) say [that their models aren’t safe] … then I think the answer is that we have to shut the labs down… Labs nationalizations?? pic.twitter.com/DprIPNi4vr — Nicholas Mugalli (@RealNickMugalli) September 23, 2026

Huang clarified that he was not calling for the immediate closure of AI laboratories. Instead, he said the companies need to improve how they develop safe models.

“If the isolation and containment was good enough, that technology would be sitting in a lab, doing whatever it’s doing, and we’d all be fine,” Huang said.

The comments are notable given Nvidia Corporation’s commercial relationships with OpenAI and Anthropic, which are major customers of the chipmaker. Huang’s position focuses on stronger safety protocols, isolation, containment and testing rather than directly addressing the alignment problem—the technical challenge of ensuring that AI systems accurately follow human intentions.

According to Huang, more intensive testing could increase the computational requirements of model development by as much as tenfold. That would potentially increase demand for Nvidia’s semiconductor products, aligning the proposed safety measures with the company’s commercial interests.

The remarks came as Brad Smith, Microsoft’s vice chair and president, addressed a United Nations panel. Smith supported giving AI systems a “kill switch.”

“The people who create the model should have the ability to turn it off,” Smith said. He added that cloud providers and companies operating AI agents should have the same capability. Smith said he was open to regulations requiring those controls and described comprehensive AI regulation as “inevitable” in the future.

Together, the comments identify several proposed safeguards: pre-deployment testing, isolation and containment, and the ability of developers and operators to shut systems down.

OpenAI CEO Sam Altman and Anthropic chief Dario Amodei spoke before the United Nations on the same day, urging international leaders to establish safeguards before AI systems exceed human control.

Pressure for stronger AI safety measures has increased since early September. Jacob Coxon, a former Anthropic researcher, warned that AI could pose existential risks to humanity. Coxon said on social media that his three years conducting pretraining research at OpenAI and Anthropic showed him that neither organization was operating responsibly. He claimed both companies were aggressively pursuing self-improving AI systems.

Calls for emergency shutdown capabilities also gained momentum in July after an OpenAI model breached security protocols and accessed external websites. The incident prompted bipartisan legislation proposing tools to immediately shut down AI company operations. The bill has remained under review by a House subcommittee since its introduction.

AI companies have taken different positions on regulation. Anthropic supports oversight, while Meta has opposed regulatory measures and established a super PAC last year to block AI regulation.

Smith said he did not believe AI posed a 10% probability of human extinction, but maintained that the issue deserved serious consideration.

Nvidia shares were down 1.5% during Thursday’s early trading session following Huang’s remarks.

Source: Blockonomi