Jensen Huang of Nvidia and Mark Zuckerberg of Meta have rejected proposals from frontier AI labs to collectively decelerate research progress in the name of safety, insisting that the associated risks should be managed independently by each company. On Tuesday, the two chief executives separately voiced their positions, emphasizing that individual firms must shoulder their own responsibilities to address public concerns that AI research could spiral out of control and pose a threat to humanity.

Huang has long dismissed such existential AI risk warnings, going further to state that the United States does not need new AI safety legislation and urging AI labs to avoid a "false dilemma" of choosing between innovation speed and developing safe products. The tech founders' remarks came after Anthropic's Dario Amodei, OpenAI's Sam Altman, and Elon Musk publicly called over the weekend for a coordinated industry slowdown to properly manage the various risks posed by AI.

This rare alignment among three typically fierce competitors emerged amid mounting researcher warnings and rising public protests that a reckless AI technology race could cause humanity to lose control over the technology, leading to catastrophic outcomes. Meta has built a major AI research lab, while Nvidia stands as the world's largest AI hardware supplier and also offers its own full suite of AI models.

Although Google and Microsoft have voiced support for cooperative industry efforts to "manage the pace of frontier technology development," Zuckerberg and Huang did not sign on. The divide reflects a deeper rift within the AI industry: on one side are leading firms like OpenAI and Anthropic that develop proprietary closed-source models, and on the other are proponents of customizable open-source models. Huang has consistently emphasized the importance of open-source models to drive widespread AI adoption, and while Meta has released some closed-source models, it broadly supports the open-source ecosystem.

Some open-source advocates argue that imposing regulation or control on research would hand an advantage to major labs that sell closed-source systems. Huang proposed on Tuesday that each company should focus on self-regulation and refrain from releasing dangerous products. "Market constraints already exist—we don't need new laws or additional regulatory rules," he said during a San Francisco event alongside Salesforce CEO Marc Benioff. Nvidia is also a significant investor in OpenAI, Anthropic, and Elon Musk's SpaceX.

The Nvidia chief said: "If you build a product or service but aren't confident in its functionality, performance, or safety, then don't release it. The principle is simple: control your own pace until you're certain what you put out will gain market acceptance." He added, "Innovation and safety can absolutely go hand in hand. Push forward as fast as possible, but if at any point you feel the company is losing control or a product fails to meet safety standards, hit pause and resolve the issue properly."

The leader of the world's most valuable company aligns with President Trump's stance against slowing AI research. On Monday, while Huang was speaking on stage at an event, he received a call from the president. During the call, Trump again dismissed the existential threat posed by AI as a "scam" and strongly criticized various efforts to slow down U.S. AI research and data center construction. Huang responded: "You're absolutely right, and we will never let that happen, Mr. President."

Later on Tuesday, Zuckerberg posted his views on X, echoing Huang's position that the responsibility for safety rests with each AI lab. "Every lab has both the responsibility and the incentive to work at the pace needed to train models safely, and they can take action on their own to achieve that goal," Zuckerberg wrote. "If a model causes harm, the lab will face significant legal liability, so they also have strong motivation to prevent such harm."

Zuckerberg revealed that Meta had delayed the launch of its Muse agent by several months, "but we didn't ask other companies to do the same." He added that expanding the pool of evaluation experts and making the evaluator community more diverse would be helpful; these independent experts are commissioned by labs to assess model performance. However, he did not respond to Anthropic's proposal to grant evaluators greater access to models.