Zuckerberg pushes back on Amodei’s call to slow AI

Meta CEO Mark Zuckerberg said AI labs already have strong incentives to keep models from causing harm, distancing Meta from Anthropic CEO Dario Amodei’s push for a global slowdown in frontier development.

In a lengthy post on X summarized by HuffPost, Zuckerberg wrote that every lab has the responsibility and incentive to move at the pace required to train models safely, and the ability to take its own actions. He argued labs face significant liability if their models cause harm.

Alignment as a race, not a pause

Zuckerberg framed trust and alignment as capabilities that will differentiate agents and models. Any lab that does not focus on alignment will fall behind, he said. That is a different theory from Amodei’s essay, which argued risk prevention needs time to keep up and that labs must slow the pace of capability gains.

Amodei’s nearly 4,000-word weekend essay drew public backing from OpenAI CEO Sam Altman, xAI CEO Elon Musk, and Google DeepMind chair Demis Hassabis. Zuckerberg’s post did not name Amodei, HuffPost noted, but the timing and substance put him on the other side of the argument.

Meta’s Muse pause as a counterexample

Zuckerberg said Meta postponed the release of its AI agent Muse to ensure safety and security. We didn’t call for everyone else to do this before we would, he added, presenting a unilateral pause as preferable to industry-wide pacing demands.

Nvidia CEO Jensen Huang offered a related line at Salesforce’s Dreamforce conference: run as fast as you can, but pause if a product is not going to be safe. Huang also put President Donald Trump on speaker after a call; Trump called extinction fears a hoax and said robots will not take over the world, according to HuffPost.

Why the split matters

The industry is arguing in public about whether competitive pressure can substitute for coordinated slowdowns. Zuckerberg’s bet is liability, reputation, and product differentiation will force labs to invest in alignment without a collective brake. Amodei’s bet is that capability races outrun those incentives unless the industry deliberately paces itself.

Meta has sounded skeptical of national AI regulatory machinery before. Politico previously reported Zuckerberg raised reservations about a national AI regulator on a call with Trump in August. This week’s post extends that skepticism into the Amodei debate without adopting the slowdown coalition’s language.