Every single on of those answers I've seen _says_ they did decode it, but each and every one of them only found the decoy message without even realizing it.
Which is probably a case of the human prompting the AI only being able to see the decoy message, so they think the AI was able to read it because it matches what they can see.
