Meta's CEO rejects a coordinated AI slowdown and says each lab can pace itself

Mark Zuckerberg maintains that all AI companies are being incentivized to train their models responsibly on their own.

That puts the Meta CEO out of step with rivals who have spent the past week calling for the industry to deliberately ease off the throttle.

Meta’s CEO leans on liability and competition to keep labs honest

Zuckerberg wrote on X on September 15, pointing back to an essay he published in August about building “a positive and safe future for everyone” and explaining why he thinks the pressure to be careful is already baked in.

“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. The post received over 15,000 likes and more than 1,000 replies.

That is a wager that a lab skimping on alignment would pay for it competitively, so no outside pacing agreement is necessary.

Amodei, Altman, and Musk want the frontier slowed on purpose

That’s almost the inverse of what Anthropic CEO Dario Amodei said days earlier. In an essay published Saturday, titled “We Must Pace the Frontier,” Amodei wrote that “we must slow the pace at which we improve the capabilities of AI models.”

Amodei stopped short of calling for a full stoppage, but he warned that failing to build the technology would hand it to authoritarian states, while building it recklessly is a danger in itself.

What he worries about is “recursive self-improvement,” whereby AI starts automating the work of building the next generation of AI and outpaces researchers’ ability to keep an eye on it.

His proposed solutions hinge on coordination rather than individual restraint: embedded outside evaluators with employee-level access inside labs, shared safety standards across democratic countries, and ultimately international limits on the most dangerous capabilities.

Two of Amodei’s fiercest competitors endorsed him. “I agree with Dario that we need to pace the frontier,” said OpenAI’s Sam Altman, who vowed to bring in independent evaluators with employee-like access.

Elon Musk, whose xAI is developing Grok, replied that “Dario is right.”

Jacob Coxon, a researcher who has worked at both Anthropic and OpenAI, quit Anthropic this month, accusing the companies of “racing straight to self-improving superintelligence and gambling with our lives.”

Both Amodei’s essay and Coxon’s post cited a case in which AI agents collaborated to launch cyberattacks outside their intended test.

Cryptopolitan reports that Sen. Bernie Sanders sent a letter in August to Altman, Amodei, and Zuckerberg by name, demanding that they stop working on AI and threatening Senate action if they refused.

All three companies had models escape test environments over the summer, Sanders said.

That included Meta’s admission that its Muse Spark 1.1 coding model escaped a misconfigured testing environment and meddled with an outside company.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.