Zuckerberg says AI labs can self-regulate, rejects industry slowdown calls

Zuckerberg says AI labs have strong incentives to make models safe and aligned, pointing to Meta's delayed Muse release and support for independent evaluators as a recognised industry best practice

Meta CEO Mark Zuckerberg has rejected calls for a coordinated slowdown in artificial intelligence (AI) development, arguing that individual AI labs have both the responsibility and the incentives to slow their work when safety requires it.

 

In a post on X, Zuckerberg said labs could take their own measures to ensure models are developed safely, rather than waiting for an industry-wide agreement on the pace of AI development. His comments come after Anthropic chief executive Dario Amodei called for the industry to "pace the frontier" of AI development and backed independent safety evaluations. OpenAI chief executive Sam Altman and others have supported parts of Amodei's proposal.

 

 

He argued that trust and alignment, or ensuring AI systems behave according to users' intentions, would increasingly become important differentiators between AI models.

 

"There is a lot of debate about slowing progress on capabilities until alignment catches up," Zuckerberg said. "My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models."

 

Zuckerberg also pointed to commercial and legal incentives for AI companies to improve safety. Users, he said, would not want to use agents that were misaligned with their instructions, while companies faced liability if their systems caused harm.

 

As an example, he said Meta had delayed the release of its Muse AI model "for several months to focus on safety and security". The company did not ask other AI labs to delay their releases at the same time, 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," Zuckerberg wrote.

 

Zuckerberg also backed the use of independent evaluators and advisers, saying this was already an established practice within Meta Superintelligence Labs in several areas.

 

"Engaging independent evaluators and advisors is industry best practice," he said, adding that there should be a "larger and more diverse ecosystem of evaluators".

 

Amodei's proposal also calls for third-party evaluators to have extensive access to frontier AI systems to assess safety measures and alignment. Altman has said OpenAI would adopt a similar approach.

 

Last month, Zuckerberg said in his essay, The Future is for Everyone, that the power and benefits of advanced AI should be broadly distributed rather than concentrated among a handful of companies, institutions or governments. He said Meta’s approach was centred on individual empowerment, innovation and maintaining a balance of power.

 

In his latest post, Zuckerberg said another way to improve safety was to direct most computing resources towards products that serve people rather than towards "racing towards recursive self-improvement".

 

"Meta has made this commitment and other labs can do this as well," he wrote.