So Anthropic caught DeepSeek, Moonshot AI and MiniMax running 24,000+ fake accounts to extract Claude's capabilities (16M+ exchanges)

Recently OpenAI had also told Congress DeepSeek is using obfuscated methods to keep distilling US models.

But honestly this was always going to happen.

Labs trained on the entire internet without asking are now upset because someone's doing it back to them through an API. The hypocrisy is hard to ignore.

> Export controls won't fix this

> You can restrict chips

> You can't restrict outputs

China will keep finding workarounds like they always do.