Meta CEO Mark Zuckerberg says he does not believe artificial intelligence will 'kill us' if developers act responsibly, rejecting calls for a coordinated industry-wide slowdown and arguing that individual AI labs should take the time needed to address safety problems before moving ahead.
In an interview with journalist Joanna Stern, recorded at Meta's Menlo Park headquarters ahead of Meta Connect 2026 and published on The New Things on 24 September, Zuckerberg was asked directly whether AI could kill humanity.
His answer was conditional: if developers act responsibly, 'then no.' He added that he was 'quite optimistic' about AI leading to 'a very positive future for everyone.'
His comments contrast with calls from some other AI executives for a coordinated slowdown in advanced AI development.
Reuters reported on 16 September that Anthropic CEO Dario Amodei had urged AI companies to slow the pace of improving model capabilities. OpenAI CEO Sam Altman and xAI's Elon Musk publicly backed Amodei's call.
Zuckerberg Says AI Labs Should Address Problems Themselves
Zuckerberg said he does not believe the AI industry needs broad coordination to avoid serious mistakes. Instead, he argued that each laboratory should decide when safety concerns require more time before development continues.
He pointed to Muse, Meta's personal AI agent, which launched in the US on 8 September. Zuckerberg said Meta already had a version that was 'quite good' months earlier but spent additional time working on privacy and security before release.
Reuters reported that Meta delayed Muse's planned April release while the company worked to strengthen its security.
Zuckerberg framed trust as a commercial requirement for AI products rather than something separate from product development.
'People aren't going to adopt it if they don't trust it,' he said.
That position echoes comments Zuckerberg made earlier in September.
Reuters reported on 16 September that he argued competition and potential legal liability give AI laboratories strong incentives to prevent their systems from causing harm. He also said Meta Superintelligence Labs uses independent evaluators in several areas, describing such evaluations as an 'industry best practice.'
Zuckerberg Says Trust and Alignment Will Matter More
Zuckerberg said AI development is entering a phase in which trust and alignment could become increasingly important alongside improvements in raw model capability. He highlighted areas including instruction-following, understanding user intent and aligning AI behaviour with a person's goals and values.
In the interview, he argued that these capabilities may increasingly determine whether people are willing to rely on AI agents.
The issue becomes more significant as AI systems move beyond answering prompts and begin acting on users' behalf.
Meta says Muse can perform tasks such as sending emails, booking travel and working across connected apps, giving the agent access to information and services people use in everyday life.
That makes trust a practical product issue as well as a safety concern for users.
Zuckerberg's argument is that this shift makes trust part of the capability race itself. Rather than treating safety and progress as opposing goals, he said labs have practical incentives to build systems that users believe will follow instructions, understand intent and handle personal information responsibly.
