Meta’s Mark Zuckerberg says AI labs don’t need to slow down: ’Alignment’ will decide who wins

The AI slowdown debate has become a centre of discussion across the tech industry, with AI leaders and researchers expressing concerns about the pace of model development. As many tech leaders, including Anthropic’s Dario Amodei, Elon Musk, and OpenAI, have joined the debate, Meta CEO Mark Zuckerberg has now weighed in, claiming that the slowdown is not necessary and that AI labs with ‘strong incentive’ should focus on building safer, more aligned models.

Why alignment matters in the AI race

Zuckerberg shared an X post on September 15, revealing his thoughts on the current AI debate. He argued that competition could drive AI labs to strengthen safeguards, as users increasingly favour systems they can trust. 

Must read: Trump calls data centre backlash a ‘hoax’ in call with Nvidia’s Jensen Huang; rules out AI slowdown debate

“There is a lot of debate about slowing progress on capabilities until alignment catches up,” he said. “My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models. Any lab that doesn't focus on alignment will fall behind.”

Zuckerberg also revealed that it delayed the release of the Muse AI model to focus on safety and security. “We didn't call for everyone else to do this before we would.”

“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 also stated that AI companies have both commercial and legal incentives that can help them build safer and more aligned models. He also said that trust and alignment could soon become among the most important factors differentiating AI agents and models.

These remarks do not go in line with what Amodei, Sam Altman, and Elon Musk have said about the need to address AI risks and safety concerns as the technology continues to advance.

Must read: AI safety warning: Sam Altman, Elon Musk back Anthropic CEO’s call to slow down frontier AI. Why?

Previously, Zuckerberg argued against US-mandated guardrails on AI slowdown. He highlighted that such mandates could impact AI development and AI labs. “Any policy that slows American model releases -- even by a month -- could add significant risk to American leadership while letting foreign models race ahead,” he said.

Meta’s four-point plan for safer AI

Meta's Chief AI Officer Alexandr Wang also added to Zuckerberg's claims by laying out a four-point AI safety approach that covers alignment, governance, liability and recursive self-improvement. 

“If we do not build models aligned with people and businesses, then they will move to more aligned options,” Wang said in the X post. He also stressed a “strong governance framework across training and deployment.” This framework should include external evaluators who keep all safety concerns in check.

Wang also highlighted recursive self-improvement (RSI), which is said to be one of the areas that could create the greatest risk of losing human control over advanced AI systems. However, he also said that Meta is not prioritising a race to develop AI that can continuously improve itself.