Zuckerberg Says AI Safety Is Each Lab's Own Job, Pushes Back on Coordinated Slowdown

Meta chief executive Mark Zuckerberg has pushed back against proposals for a coordinated slowdown in advanced artificial intelligence development, arguing that responsibility for keeping the technology safe rests with individual companies rather than a collective industry pause.

His comments, posted on social media, came days after Anthropic CEO Dario Amodei outlined a three-step plan for AI safety that included a proposal for companies and countries to coordinate a slowdown in development. The idea drew quick backing from other technology leaders, including OpenAI's Sam Altman and Elon Musk at SpaceXAI.

Zuckerberg did not name Anthropic or any other company in his post. He said AI firms already have both the motivation and the obligation to build their models safely, pointing to what he described as significant liability they face if their systems cause harm.

"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," he wrote.

As evidence of that approach, Zuckerberg said Meta held back the release of its Muse AI agent for several months to ensure its safety and security. "We didn't call for everyone else to do this before we would," he said.

He also urged companies to focus on building safe AI models rather than systems capable of recursive self-improvement, in which a model can automatically improve itself. Directing the bulk of computing resources towards serving users, rather than racing towards that capability, is among the best ways to develop the technology safely, he wrote.

The exchange reflects a widening debate among industry leaders over how to restrain advanced AI, amid warnings from researchers about the technology's potential threats to humanity.