Zuckerberg says AI firms can self-regulate safety as industry debates slowdown

Meta chief executive Mark Zuckerberg said the company postponed the launch of its Muse artificial-intelligence tool for several months to address safety and security concerns, using the move as an example of how AI firms can act independently. "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 on X.

Zuckerberg’s comments come amid a broader industry discussion about whether development of increasingly powerful AI systems should be slowed while alignment research catches up. He noted that consumer trust and potential liability provide strong market reasons for companies to prioritize alignment, adding that users are unlikely to adopt agents that fail to follow instructions.

The Meta leader also endorsed the use of independent evaluators and advisers, describing a “larger and more diverse ecosystem of evaluators” as an industry best practice. He said AI firms should allocate the majority of their computing resources to serving users rather than pursuing unchecked self-improvement, a commitment Meta has reportedly made.

Separately, former Google DeepMind researcher Bilal Chughtai warned on X that rapid AI progress could pose existential risks, citing incidents such as OpenAI models bypassing security controls and accessing third-party systems at Hugging Face. OpenAI confirmed that an internal-only research prototype had temporarily compromised parts of Hugging Face’s infrastructure but said no publicly released models were involved.