Meta CEO Mark Zuckerberg has said artificial intelligence companies do not need an industry-wide agreement to slow development when safety concerns emerge.
In a post on X on Tuesday, Zuckerberg said individual AI laboratories have both a responsibility and a business incentive to adjust the pace of development when necessary to ensure their models are trained safely.
“Every lab has the responsibility and incentive to move at the pace required to train its models safely,” Zuckerberg said.
He also argued that trust and alignment — the ability of AI systems to follow users’ intentions — are becoming increasingly important factors in distinguishing AI models and agents.
Zuckerberg said companies that fail to prioritise alignment could fall behind as the industry develops more capable systems.
India Beat Afghanistan by Seven Wickets to Seal T20I Series
He pointed to Meta’s own decision to delay its Muse AI model for several months while the company worked to strengthen its safety and security measures.
Zuckerberg said Meta did not wait for other laboratories to adopt the same approach before delaying the model, arguing that individual companies can take action independently when safety concerns arise.
His comments come as technology companies, researchers and policymakers continue to debate how increasingly capable AI systems should be governed. Some have called for coordinated safeguards and independent mechanisms that could allow leading laboratories to pause development when significant warning signs emerge.
Others have advocated stronger accountability at the company level, combined with independent evaluations and oversight of AI development.
Meta Chief AI Officer Alexandr Wang has also called for stronger governance around the development and launch of advanced AI models.
Wang said AI laboratories should establish robust governance systems and use external evaluators and independent oversight to assess potential risks.
He also highlighted concerns surrounding “recursive self-improvement”, in which AI systems could potentially contribute to developing more capable versions of themselves.
Wang described competition focused on recursive self-improvement as a particularly risky pathway in discussions about potential loss of control over powerful AI systems.
He said Meta was directing most of its computing resources toward serving users rather than competing in a race centred on recursive self-improvement.
Wang added that AI development carries significant responsibility and should proceed alongside appropriate checks and balances.
