Zuckerberg, Huang reject coordinated AI slowdown push

Meta Chief Executive Mark Zuckerberg and Nvidia CEO Jensen Huang have pushed back against calls for coordinated limits on frontier AI development, arguing that companies already have strong incentives to make their systems safe.

Their comments sharpen a debate that has widened since Anthropic CEO Dario Amodei called for AI companies to slow capability gains when safety work cannot keep pace. OpenAI CEO Sam Altman has backed greater industry coordination, while stopping short of endorsing every element of Amodei’s proposal.

“There is a lot of debate about slowing progress on capabilities until alignment catches up,” Zuckerberg said Tuesday. “My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models.”

Zuckerberg said market incentives already push developers toward safer systems because users will avoid models they cannot trust and companies face liability when their products cause harm.

“Any lab that doesn’t focus on alignment will fall behind,” he said.

He pointed to Meta’s decision to delay the release of its Muse personal AI agent for several months while it worked on safety and security.

“We didn’t call for everyone else to do this before we would,” Zuckerberg said. “We just did it as part of our day-to-day work because it was clearly the right thing for people and for us.”

Meta launched Muse on September 8, describing it as a personal AI agent built with separate security controls and approval requirements for sensitive actions.

Huang made a similar case at Salesforce’s Dreamforce conference in San Francisco, saying safety should be treated as a core engineering responsibility.

“Safety is paramount in a lot of ways. It’s job one,” Huang said. “However, safety is an engineering problem.”

He said companies should not release systems they believe are unsafe but rejected the idea that speed and safety necessarily conflict.

“Innovation, speed, and safe products, it’s a false choice,” Huang said. “You could definitely have both at the same time.”

“So run as fast as you can,” he added. “But if you feel at any given point in time the company’s out of control, or the product’s not going to be safe, take a pause and make sure you get it right.”

Amodei is arguing for a more coordinated approach. In an essay, titled We Must Pace the Frontier, he said AI capabilities are advancing faster than safety work and proposed a three-stage framework involving embedded third-party evaluators, industry-wide coordination and eventually international cooperation.

He stressed that pacing did not mean halting model training or technical progress, but giving developers and outside evaluators enough time to assess increasingly capable systems.

Altman has also backed coordination among leading developers.

“I think it’s great for our industry to say we want to come together and we want to be able to coordinate and make sure we have enough time to do this safely,” he said at Dreamforce.

Altman said public concern grows when commercial or geopolitical competition creates doubts over whether companies or countries will put safety first.

The disagreement is therefore less about whether AI systems should be safe than about how that safety should be enforced.

Amodei wants common commitments that reduce the pressure on individual companies to race ahead. Zuckerberg and Huang argue that developers can make those decisions themselves, including slowing or withholding products when necessary, without requiring their competitors to do the same.