Anthropic co‑founder warns China is biggest obstacle to an AI pause

Dario Amodei, the co‑founder of Anthropic, told CNBC that China presents the most formidable challenge to his effort to slow the pace of advanced artificial‑intelligence development. He argues that without Beijing’s participation, any voluntary slowdown by Western firms would be ineffective.

Amodei’s proposal calls on leading AI companies to voluntarily limit the size and speed of new models, citing safety and societal risks. According to CNBC, the idea has earned public support from OpenAI chief Sam Altman, DeepMind chair Demis Hassabis and entrepreneur Elon Musk, who all praised the need for a measured approach.

China’s AI sector, however, is moving at a breakneck speed. The Chinese government has set a goal to dominate key AI technologies by 2030, and domestic labs are scaling models that rival those of U.S. rivals. This creates a strategic dilemma: a unilateral slowdown by American firms could hand a competitive edge to Chinese players, undermining the very safety goals the pause seeks to protect.

The debate over an AI slowdown is not new. Since 2023, industry leaders and policymakers have raised alarms after high‑profile incidents involving disinformation, deepfakes and unintended model behavior. The United States and the European Union have floated regulatory ideas, but a binding international framework has yet to materialise, leaving self‑regulation as the primary lever.

If China continues to accelerate while U.S. companies curtail development, the gap in capabilities could widen, influencing market share, talent migration and geopolitical leverage. Conversely, a truly global pause could temper the rapid emergence of potentially hazardous AI systems, but only if all major players, including Beijing, commit.

Amodei plans to convene a summit of AI CEOs later this year to flesh out concrete terms for a pause. He stresses that any successful agreement must be inclusive, otherwise the AI landscape risks fragmenting into competing camps with divergent safety standards.