I’m sure they do, it’s not like anyone can stop them. That’s valuable training data after all, and reverse engineering with LLMs is easy because it has concrete goals (compilation must be 1:1 with the original binary).