Zuckerberg urges responsible AI development, stresses balance of power

Meta chief executive Mark Zuckerberg has weighed into the intensifying debate over the pace of artificial intelligence development, arguing that safety and alignment should be treated as competitive strengths rather than obstacles to progress.

In a post on X, Zuckerberg pointed to a blog post he had written last month and said every laboratory has both the responsibility and the incentive to train its models at a pace that keeps them safe. He argued that users will not adopt agents that are misaligned with them or that fail to follow instructions, giving labs a natural reason to improve alignment.

"Trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models," he said, adding that any lab that neglects alignment will fall behind. He also noted that labs face significant liability when their models cause harm, creating a further incentive to prevent misuse.

Zuckerberg cited Meta's own decision to delay shipping Muse for several months to focus on safety and security, saying the company did so as part of its routine work rather than asking others to follow first.

His position overlaps with that of Anthropic CEO Dario Amodei and Microsoft's chief executive on the value of independent evaluation of AI systems. Zuckerberg said engaging independent evaluators and advisors is industry best practice and that his company already does so in several areas, adding that a larger and more diverse ecosystem of evaluators would help the field.

On the question of how compute should be used, he said committing the significant majority of it to serving people, rather than racing towards recursive self-improvement, is among the best ways to develop the technology safely. Meta has made that commitment, he said, and other labs can do the same.

"I believe the key to building a positive future for everyone is maintaining the right balance of power," he wrote, describing this as within the industry's power to achieve.

In his earlier blog post, Zuckerberg framed access to superintelligence as a defining question of the age, asking whether it will be centralised among a few institutions or serve as a tool that empowers everyone. He proposed a philosophy built on individual empowerment as the source of prosperity, invention as the primary purpose of superintelligence, and balance of power as the foundation of safety.

He said superintelligence is likely to be among the most important technologies in history, with opportunities and challenges greater than any seen in recent lifetimes, and that these should be taken seriously. At the same time, he pushed back against doom-laden narratives, questioning why anyone who believes AI will eliminate most jobs would rush to build such a future, and describing the idea that extreme concentration of power is the only safe path as inherently problematic.

Meta, he said, is committed to building on the principles of individual empowerment, invention as AI's purpose, and a balance of power favouring people as the basis for addressing safety risks.

The debate over slowing AI development has been described as an inflection point for the industry. It began with questions of principle raised after Anthropic researcher Jacob Coxon publicly resigned, claiming that building superintelligent systems poses a genuine risk to humanity, and has since grown into a wider argument over how the AI industry should evolve.