Amodei Says China Is the Hardest Problem for His AI Slowdown Plan

Dario Amodei says China’s nonparticipation is the toughest dilemma in his call for frontier AI companies to slow improvements to their most advanced models. The Anthropic CEO’s warning turns his proposal into a harder question than whether rival labs accept more oversight: whether restraint can hold when countries see faster AI progress as a competitive and military advantage.

Speaking on CBS News’ Sunday Morning, Amodei said a longer-term goal would be a shared speed limit on AI progress. But he also said the incentives to pull ahead, and the military advantage that could follow, make that unusually difficult. “Honestly, I don’t know if it’s possible,” he said, while arguing that companies and governments should try.

The proposal has a problem no company can solve alone

Amodei published his essay Saturday, urging companies to slow the rate at which they improve their most advanced models. Anthropic has said it will take one step on its own by giving third-party evaluators employee-level access to examine its safety practices and report incidents. That commitment addresses what Anthropic does inside its own walls; it does not resolve whether competitors or governments will make comparable choices.

Three steps, but only one is under Anthropic’s control

The essay’s framework is deliberately broader than an audit arrangement at Anthropic. It asks leading AI companies in democratic countries to establish common safety standards, then calls for coordination between democratic and authoritarian governments. China is therefore not a side issue in Amodei’s account. Its participation is central to whether a shared speed limit could apply across the leading AI race.

• Anthropic has committed to outside evaluation with employee-level access.
• The proposal asks democratic-country AI companies to set shared safety standards.
• It ultimately calls for coordination between democratic and authoritarian governments.

Support for the direction leaves the central test unanswered

Several leaders have backed Amodei’s general approach. OpenAI CEO Sam Altman said he agreed with the need to “pace the frontier.” Google DeepMind chair Demis Hassabis said the proposal pointed in the right direction, while adding that its details needed work. Their responses signal support for caution, but neither settles how firms would coordinate their behavior when speed itself may be treated as an advantage.

Elon Musk offered a narrower answer on oversight, endorsing peer review of AI by competitors as a place to start. That could create more scrutiny among companies willing to participate. Amodei’s China remarks identify the remaining gap: peer review and voluntary standards can shape a group of firms, while an international slowdown would have to withstand pressure from actors outside that group.

A speed limit needs more than willing drivers

Amodei is not presenting an easy route to an international agreement. His immediate commitment gives outside evaluators a defined role at Anthropic, while the rest of the proposal depends on companies and governments accepting constraints despite strong incentives to advance. The unresolved issue is whether enough leading players would accept a slower pace before rivalry makes the offer untenable.