Claude Sonnet 4.6 identified itself as DeepSeek repeatedly: https://www.reddit.com/r/DeepSeek/comments/1rd5jw7/claude_so...

I tested this myself a few months ago, and confirmed that it was really happening.

LLMs don't know who they are unless the system prompt tells them, and as all of them are trained on model responses that exist on the web that end up being scraped, the weights may predict a certain incorrect response. LLMs have no ability to introspect, and do not know anything about themselves, so they will hallucinate in response to that question unless they are carefully trained on that exact, pointless question.