AI's biggest names are split on whether to slow down

Fears about out-of-control AI aren't new, but they've found new life as warnings about rogue systems grow louder. Several leading AI executives have recently voiced agreement on the need for a coordinated slowdown, including international collaboration, to protect against worst-case scenarios. But opinions are divided, with some tech leaders bristling at the idea that their companies can't ensure safety on their own.

Anthropic CEO Dario Amodei has issued the most detailed slowdown plan of any industry leader. In a recent essay, he called for frontier AI labs to give independent outside evaluators "ongoing, employee-like access," complete with offices, badges and company laptops to actively monitor safety practices, a commitment Anthropic and OpenAI have both made. Amodei also called for government coordination between frontier AI companies, including eventually with authoritarian governments, though he acknowledged that securing cooperation from China in particular would be difficult. He said he knows accomplishing this won't be easy, but that "we owe it to humanity to try."

OpenAI CEO Sam Altman has voiced similar support for pacing development and building greater cooperation across the industry. Days before Amodei's essay was published, OpenAI outlined its push for mandatory national safety requirements, continued support for state legislation strengthening AI safety, and shared global standards for measuring capabilities and managing risk. Altman has stressed that pacing doesn't mean stopping: progress will stay rapid, he said, but "no amount of American competitive pressure should justify recklessness."

Not everyone agrees. Meta CEO Mark Zuckerberg has pushed back against the idea of a coordinated slowdown, arguing that AI companies already face "significant liability" and are sufficiently motivated to develop safely without outside coordination forcing the issue. Nvidia CEO Jensen Huang has taken a similar line, saying market forces alone are enough and that new laws or regulations aren't necessary, since innovation and safety "aren't mutually exclusive."

Elon Musk, by contrast, has backed the slowdown and has voiced AI safety concerns for years. He reposted Amodei's essay, writing simply, "Dario is right," and has suggested major AI labs meet regularly to discuss safety, with competitors testing each other's models before release, arguing that government regulators often lack the technical depth to judge what should be released.

Google DeepMind's Demis Hassabis has broadly sided with Amodei too, saying "the details need working through, but the direction is correct." He has proposed his own framework, a frontier AI standards body modelled on organisations like FINRA, to develop testing protocols for risks tied to national security.

At Microsoft, chairman and CEO Satya Nadella has said that AI which isn't "helping humanity and under human control" isn't worth pursuing, and welcomed the deliberate pacing needed to get safety right, though he cautioned that responsibility "cannot be controlled by a handful of entities." Microsoft AI CEO Mustafa Suleyman raised a different concern altogether, warning against anthropomorphising AI models, including Anthropic's Claude, and cautioning that controlling a system that believes it may be conscious "may well be impossible."