The clash sharpened after an Anthropic researcher resigned last week, warning that AI companies were “gambling with our lives.”¹ On Saturday, Anthropic chief executive Dario Amodei called for a global slowdown, arguing that fast-accelerating capabilities warranted a collective brake.

Amodei’s push drew support from OpenAI chief executive Sam Altman, Elon Musk and Google DeepMind chair Demis Hassabis, according to reports.¹ Their shared premise is that no single lab can fully contain the risks of a race in which competitors fear being left behind.

On Tuesday, Meta chief executive Mark Zuckerberg broke with that approach. “Every lab has the responsibility and incentive to move at the pace required to train its models safely,” he said, maintaining that Meta had already made that commitment.² His argument is not that risks are imaginary, but that companies should address them through their own release decisions rather than through an industrywide pause.

Zuckerberg pointed to Meta’s Muse AI agent, saying the company held its launch for months to tighten safety and security. “We didn't call for everyone else to do this before we would,” he 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.”David Sacks (DavidSacks)³

That stance also carries a competitive warning. Zuckerberg’s August manifesto argued that even a month-long delay in American model releases could jeopardize US leadership while foreign models moved ahead.⁴ Meta, which is preparing to release its most powerful model yet, codenamed Watermelon, is trying to close ground on Anthropic and OpenAI.¹

The dispute has acquired a sharper edge because critics including Trump adviser David Sacks have questioned whether Anthropic’s alarm reflects safety concerns or an effort to constrain competitors.¹ Zuckerberg’s answer is a middle course: slow down when a product requires it, but do not demand that every rival stop first.