Meta CEO Mark Zuckerberg has rejected calls for artificial intelligence companies to coordinate a slowdown in developing advanced AI, arguing that every company should take responsibility for determining how quickly it can safely train and deploy its models. His comments came after Anthropic CEO Dario Amodei proposed a three-step approach to AI safety that included coordinated action by companies and governments to slow development if necessary, an idea that drew support from technology leaders including OpenAI CEO Sam Altman and Elon Musk.
Safety responsibility
In a social media post, Zuckerberg said AI developers were already strongly motivated to ensure their systems were safe because they could face significant liability if their technology caused harm. He argued that each laboratory should be able to adjust the pace of development according to the safeguards required for its own models rather than waiting for the entire industry to act collectively. Zuckerberg did not directly name Anthropic or other companies while making his argument. The disagreement reflects a broader debate within the technology industry over whether increasingly powerful AI systems require coordinated restrictions or whether companies can manage emerging risks through their own safety frameworks.
Meta delay cited
Zuckerberg pointed to Meta’s decision to postpone the release of its Muse AI agent for several months while additional safety and security work was carried out. He said Meta did not demand that competing companies halt their own development during that period, presenting the episode as an example of how individual developers could delay products when their internal assessments indicated that more safeguards were needed. His position contrasts with Amodei’s proposal for greater coordination among leading developers and governments as concerns grow over whether future AI systems could develop capabilities that become increasingly difficult for humans to control.
Self-improvement warning
While opposing a coordinated slowdown, Zuckerberg cautioned companies against prioritising AI systems designed to improve their own capabilities automatically, a process generally described as recursive self-improvement. He argued that most computing resources should instead be devoted to operating AI products and services that directly benefit people. Directing the “significant majority” of computing capacity towards serving users rather than competing to achieve recursive self-improvement would, he said, be one of the strongest ways of keeping advanced AI development safe.
Industry debate grows
The exchange comes as leading AI developers and researchers increasingly debate how to manage risks associated with rapidly advancing models. Some have argued that voluntary company-level safeguards may prove insufficient if competitive pressure encourages developers to move faster, while others contend that broad restrictions could unnecessarily slow useful technological progress. Zuckerberg’s intervention places Meta on the side of company-specific safety decisions, while acknowledging that developers have a responsibility to slow their own work when they determine that additional safeguards are necessary.