Anthropic spent years marketing itself as the most responsible artificial intelligence company - the one that loudly warns how dangerous its models are. Last Friday that strategy came back as a boomerang: the U.S. government ordered an emergency worldwide shutdown of the company's most powerful model, at exactly 5:21 p.m. Eastern time.

Two of the newest models are affected - Claude Mythos 5, available to only about 50 vetted organizations, and the just-released Claude Fable 5 with public access. The official justification is export control aimed at foreign nationals, prompted by an alleged "narrow, non-universal" security breakthrough that let the model read code and uncover software weaknesses.

Anthropic strongly disagrees. The company claims the same capability already exists at competitors - among others at GPT-5.5 from OpenAI - and that cybersecurity professionals use it routinely. In other words: the company punished is the one that's first to say something might be dangerous, while the same thing runs freely at the others.

The irony is complete. CEO Dario Amodei builds the entire marketing story on how exceptionally powerful, and therefore exceptionally risky, the Mythos models are. The rival Sam Altman of OpenAI previously called this "fear-based marketing." Now that fear has become a regulatory pretext - and struck directly at Anthropic's plans to go public.

A lesson every tech company will now remember: if you publicly spin the story that your product is too powerful to be safe, don't expect only applause. Sometimes that's exactly what attracts whoever has the power to switch it off. In the Balkans we've known this for ages - the one who brags most about his strength is the first they put in the crosshairs.