Of course it believes the proof is sound, it wrote it. If you want to check an LLM's output, you should use a different LLM.
Use a human maybe.
Only people can really verify clankers.
Can't trust anything LLM since it will confidently lie too.
It can't take responsibility for verification so it can't verify.
Your comment is not substantiated at all.