That feels plausible to me.
You can trigger this type of issue by ChatGPT then reading the transcript.
The model doesn’t know you interrupted it, so continued assuming he had heard the steps.