Mark Zuckerberg Says AI Labs Can Build Safely As Yet Another AI Expert Warns Of Extinction

Artificial intelligence (AI) companies are facing calls to slow development over fears that increasingly capable systems could become difficult to control.

Meta CEO Mark Zuckerberg, however, argues AI labs already have competitive and legal incentives to prioritize safety, even as researchers warn the consequences of getting it wrong could be existential.
 

TL;DR

 

• Zuckerberg says competition, alignment, and potential liability give AI labs strong incentives to develop systems safely.
• Meta previously delayed its Muse AI agent to strengthen security and uses independent evaluators for AI safety.
• Former Google DeepMind researcher Bilal Chughtai says the AI race could create extreme risks and is calling for coordinated action.
• Anthropic researcher Evan Hubinger puts the risk of AI killing humanity within a decade at more than 10%.

 

Zuckerberg Says Competition And Liability Can Push AI Labs Toward Safety

While some AI leaders are pushing for companies to collectively slow the development of more capable systems, Zuckerberg believes individual labs already have enough reason to act cautiously.

"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 said, arguing that companies neglecting alignment would ultimately fall behind competitors.

He also pointed to potential liability if AI models cause harm. Meta itself delayed the release of its Muse AI agent earlier this year to improve security rather than waiting for competitors to make similar commitments.
 

 

Zuckerberg added that Meta Superintelligence Labs already works with independent evaluators in several areas, describing the practice as something other AI companies could adopt independently.

The company has also committed the "significant majority" of its computing capacity to products serving immediate user needs instead of focusing primarily on AI systems capable of recursively improving themselves.

Yet that confidence in company-level safeguards is arriving just as researchers are escalating their warnings about what an uncontrolled AI race could mean.
 

 

Former DeepMind Researcher Warns AI Race Could Threaten Humanity

Bilal Chughtai, a former Google DeepMind research engineer who worked on AGI safety and alignment before leaving the company in July 2026, has become one of the latest researchers to raise an existential warning.

"I earnestly believe that AI has the potential to kill us all, and that we might be running out of time to avoid this outcome," Chughtai wrote.

His warning follows similar concerns from Anthropic researchers. Evan Hubinger said he believes there is a greater than 10% chance AI could kill all humans within the next decade, while Anthropic CEO Dario Amodei has called for advanced AI development to slow down.

Chughtai argued that safely navigating AI development remains possible but requires coordination between companies rather than what he called a "manic race."

"We need to pace AI development to a speed that society can handle," he said.

That leaves the AI industry confronting a widening divide over the same question: whether competitive incentives can make increasingly powerful AI safer, or whether competition itself is becoming the risk.