There are already published examples where there was very specific info and guidance provided.

Beyond the info and guidance, there's also the classic sycophantic encouragement. Humans are allowed to publish the Anarchist's Cookbook, but they tend to get into trouble when it becomes "based on your manifesto, I would suggest the following targets". Of course, AI isn't human, and treating a software program like a human probably isn't good law, but OpenAI are very keen to suggest it's legally equivalent to a human when it suits them...