Maybe another AI could use the proof.
How would we know it was correct?
If you feed an AI nonsense in its training data, it will generate nonsense