Because they want the fixes that B-ZZZ learned about and they may not be able to avoid letting the model know that it made an error, unless they suddenly go silent to the model about what happened.
New job under AI. Go work for company A, but use it to write programs that use Company B's stack, but make sure to overcomplicate everything and "correct" the LLM into doing the wrong thing. Make sure Company B gets the results of your "improvements".