Anthropic's repeated AI risk warnings may have triggered US export ban on its latest models

According to an FT analysis, Anthropic used risk-related language five times per 1,000 words in 2026, far more than OpenAI's 0.6 per 1,000. Last week, Washington barred foreign nationals from using Anthropic's new models, Mythos and Fable. Some technologists argue that the $965 billion AI company's persistent warnings about AI dangers, especially concerning Mythos, influenced the decision.