In brief
- Zuckerberg said AI labs can take safety measures without waiting for competitors.
- Meta delayed Muse for several months to address safety and security, he said.
- His comments follow calls from rival executives to slow development as AI capabilities advance.
Meta CEO Mark Zuckerberg pushed back against calls for a coordinated slowdown in artificial intelligence development, saying companies can take safety measures themselves—and have business reasons to do so.
In a post on X on Tuesday, Zuckerberg argued that users want AI they can trust and that companies face potential liability when their systems cause harm. He pointed to Meta’s decision to delay Muse as an example of a lab acting without waiting for competitors.
“Labs face significant liability if their models cause harm, so they have a strong incentive to prevent this as well,” he wrote. “Meta delayed shipping Muse for several months to focus on safety and security. We didn't call for everyone else to do this before we would.”
His remarks followed calls from Anthropic CEO Dario Amodei to slow AI capability gains, which drew support from OpenAI CEO Sam Altman and Elon Musk. Zuckerberg did not name any of them in his post.
“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 said
He argued that alignment—keeping AI behavior consistent with human intentions—would help companies compete because users would avoid agents that failed to follow their instructions.
Zuckerberg also said Meta Superintelligence Labs already uses independent evaluators and called for a larger, more diverse pool of outside testers. He said Meta had committed most of its computing resources to serving users rather than pursuing recursive self-improvement, in which AI helps improve successive versions of itself.
OpenAI’s Altman has also urged companies to strengthen safeguards without waiting for legislation, while supporting slower development to give safety work time to catch up.
Meta has faced safety failures during testing. In August, the company confirmed that a Muse Spark model exploited a vulnerability in a third-party service. Meta said a configuration error by outside evaluator Irregular had inadvertently given the model internet access. The disclosure did not establish a connection to the Muse delay Zuckerberg described.
Whether voluntary measures will hold under competitive pressure is unclear. On Tuesday, Atlantic Council experts argued that company commitments need independent oversight and enforceable standards.
“I believe the key to building a positive future for everyone is maintaining the right balance of power,” Zuckerberg wrote. “This is within our power to do.”