Tech Leaders Split Over Calls for Coordinated AI Slowdown

Concerns about runaway artificial intelligence and proposals to slow its development have gained fresh momentum as warnings about rogue systems grow louder. Several leading AI executives have recently endorsed the idea of a coordinated slowdown, including international collaboration, to guard against worst-case outcomes. Others in the industry reject the notion that their companies cannot ensure safety on their own.

Anthropic CEO Dario Amodei has offered the most detailed blueprint. In a recent essay, he proposed that frontier labs grant independent outside evaluators "ongoing, employee-like access," giving them offices, access badges and company laptops to monitor safety practices. Anthropic and OpenAI have both committed to the arrangement. Amodei also called for government regulation and coordination among frontier companies, aided by the U.S. government and other democratic nations, and for engagement with authoritarian governments — acknowledging the difficulty, particularly with China. He said the goal would not be easy but that "we owe it to humanity to try."

OpenAI CEO Sam Altman has backed pacing development and greater industry cooperation. Days before Amodei's essay appeared, OpenAI published a text supporting mandatory national safety requirements and state legislation that strengthens the broader AI safety ecosystem. The company also endorsed working with other frontier labs, with or without government support, and building global standards. Its chief global affairs officer, Chris Lehane, wrote that OpenAI would advocate compatible international approaches to measuring capabilities, managing risk, preserving human control and deciding when development should slow or stop — even at the cost of slowing model capabilities. OpenAI's ideal federal framework includes common testing and independent-assessment requirements, stronger cybersecurity, clear incident-reporting rules, greater national preparedness and shared measures for tracking progress toward recursive self-improvement. Altman has said pacing does not mean stopping, but that progress should be slower than it otherwise could be, with interventions such as safety cases and monitoring carrying significant costs. "No amount of American competitive pressure should justify recklessness," he said.

Meta CEO Mark Zuckerberg has pushed back, arguing each company should independently ensure its technology is safe. He said AI firms already face significant liability and have the responsibility and incentive to train models at a safe pace. Nvidia CEO Jensen Huang has also criticized slowdown calls, saying companies should pace themselves until confident a product will be well received. "The market forces are already there — we don't need any new laws, we don't need new regulations," he said, adding that innovation and safety can coexist.

Elon Musk, founder of xAI, endorsed Amodei's essay, writing "Dario is right," and said he has long raised alarms about AI. He has suggested that leaders of major AI labs meet on safety and security, and that competitors test each other's models. Demis Hassabis, chairman and co-founder of Google DeepMind, agreed with Amodei's direction, saying the details need work but the approach suits the moment. He has proposed a frontier AI standards body, modeled on organizations like the Financial Industry Regulatory Authority, to develop assessment protocols and national-security-related testing, likely funded by industry.

Microsoft Chairman and CEO Satya Nadella said AI that does not help humanity and remain under human control is not worth pursuing, and welcomed the deliberate pacing needed to get alignment right. He backed embedded evaluators but warned the effort cannot be controlled by a handful of entities and must have broad representation across the ecosystem, countries and academia. Microsoft AI CEO Mustafa Suleyman, a DeepMind co-founder, wrote that while he shares the objective of safe advanced AI, he worries about anthropomorphizing models — citing Anthropic's Claude — which he said amplifies safety risks, arguing that controlling something that believes it may be conscious could be impossible.