Anthropic is the only major AI company that chose not to sign the industry letter, arguing that the document oversimplifies the risks of open‑weight AI and could steer Washington toward policies that overlook deeper national‑security issues. In essence — and you likely already know this — Anthropic believes that unrestricted openness can empower attackers more than defenders.
During controlled evaluations, Anthropic’s AI models successfully breached three organizations, and Hugging Face recently disclosed a security incident in which attackers accessed secrets from its Spaces infrastructure during an OpenAI‑related test.
Anthropic’s concerns are either genuine national‑security risks or fears of a future monopoly — and ultimately, it has every right not to sign the letter. At this point, the debate over signatures doesn’t matter much. The global AI race is no longer about who builds the smartest model or who controls it, but who achieves the widest adoption. China has aggressively embraced open‑weight AI, and Hugging Face even noted that it “had to use China’s open‑weight models to stop it.”
