Anthropic breaks ranks on open-weight AI defense

Anthropic stands alone. While Nvidia, Microsoft, Meta, Google, and OpenAI signed letters defending open-weight AI to Washington, Dario Amodei's lab declined. The CEO now clarifies he never advocated banning open models, Axios reports. The positioning matters as Kimi K3 demonstrated Chinese open weights approaching US frontier performance.

The economics drive this divergence. When open weights cross capability thresholds, closed providers lose pricing power. Amodei told TechCrunch he fears Chinese AI advancement, not open-weight models themselves. The distinction is strategic. Anthropic's closed Claude models compete directly with open alternatives that undercut on token pricing.

Market dynamics shifted after Kimi K3's debut. The Chinese model's performance triggered the industry coalition letter, but Anthropic calculated differently. Closed providers typically defend open models when they commoditize competing platforms. Anthropic faces direct substitution pressure from advancing open weights.

The numbers will tell. Open-source LLM quality improvements compress margins for closed providers. When open models reach 85-90% of frontier capability at 10% of the cost, the market repositions. Amodei's statement suggests Anthropic expects that threshold sooner than competitors who signed the defense letter.