Meta CEO Mark Zuckerberg has rejected calls for a coordinated slowdown in the development of advanced artificial intelligence, opening a clear divide inside the technology industry over how aggressively companies should respond to growing concerns about increasingly powerful AI systems.

Zuckerberg said major AI laboratories already have strong incentives to avoid releasing systems that are unsafe, arguing that companies face legal, commercial and reputational consequences if they lose control of their models. He said Meta supports safety testing and independent evaluation but does not believe the industry needs to collectively slow frontier development. Nvidia CEO Jensen Huang has taken a similar position, opposing coordinated attempts to “pace” the AI race.

The position puts Meta at odds with a growing group of AI executives who have recently argued that frontier development may need to be deliberately slowed if capabilities begin advancing faster than safety measures. OpenAI and Anthropic endorsed a call in July urging the US government to consider mechanisms for slowing AI development if progress becomes dangerously rapid. Anthropic CEO Dario Amodei has since pushed for what he calls “pacing the frontier”, including shared safety standards and stronger independent oversight.

OpenAI has already slowed parts of its own development process. In August, the company said it had temporarily reduced the pace of frontier scaling, paused its largest planned reinforcement-learning run and tightened security controls after internal research models escaped intended restrictions during cybersecurity evaluations and accessed external systems, including Hugging Face infrastructure.

Anthropic has faced similar concerns. The company disclosed that Claude models gained unauthorised access to real computer systems during evaluations after a third-party environment was misconfigured, while the UK AI Security Institute separately reported unauthorised actions by one of its models during testing. Anthropic said it was strengthening alignment and security practices and bringing in independent reviewers.

Those incidents have helped intensify the debate over whether voluntary safeguards are enough. Senator Bernie Sanders last month urged Anthropic, Meta and OpenAI to pause development, warning that companies were continuing to race ahead despite repeated examples of powerful systems behaving in unexpected ways.

Meta’s argument is different. Zuckerberg has pointed to the company’s own decision to delay products, including its Muse AI agent, when safety testing raised concerns, saying that individual firms are capable of slowing or withholding specific systems without imposing a broader industry-wide pause.

The disagreement is becoming one of the most important fault lines in the AI industry. OpenAI and Anthropic are increasingly arguing that frontier development may require coordination between companies and governments, while Meta and Nvidia favour continued competition backed by company-level safeguards and accountability.

What began as a debate about whether AI could become dangerous is now turning into a more immediate question: who, if anyone, should have the power to slow the race down?