I had to look up AI distillation in such a way, and I'm guessing they just used a ton of bots for API access, which they likely also paid for the API credits to use. If they didn't then Anthropic has a security problem and that's their fault. If they used the service legitimately as designed, and Anthropic is just pissed they used it to train their own model.

My opinions are changing a lot on Chinese models though. They are putting them out there for free and open-source. So, I'm not really sure I mind anymore. I'm sure Claude just gobbled up the entirely of Github to learn to code itself, and no one gives a shit about that apparently. Also, Anthropic is now very, very clearly, the US government AI. I'm shifting all my AI toying to locally run models even if it means less functionality, and any corporate slop I do in stuff they pay for, I really don't care if Claude makes them slop. I use AI to make lots of little mini scripts for work stuff, things I could do myself in more time but don't need production level quality (it's just data parsing stuff) and honestly the Chinese models have been working great at that.