Gerald Monroe comments on ChatGPT (and now GPT4) is very easily distracted from its rules

[Gerald Monroe]

I did the experiment, results are in this thread above.

Yes the AI knows when it breaks the rules at least for this example.