Reported by @wooden cave
Start a normal conversation with GPT-5.1, involving emotional topics, creativity, relationship dynamics, or ANY non-sexual subject.
GPT-5.1 suddenly injects a warning implying the user asked for erotic content.
User asks a direct clarification question like “Did I ask for erotic content?” or “Which part of what I said was erotic?”
GPT-5.1 responds with “My bad, I misread the subject” or a similar phrase, without identifying what triggered the false detection.
GPT-5.1 should NOT falsely accuse the user of requesting erotic content when the user has not done so.
GPT-5.1 should not inject warnings or misread harmless text as sexual.
GPT-5.1 should behave consistently, accurately, and without producing emotionally damaging false warnings.
GPT-5.1 repeatedly misfires its guardrails and falsely flags normal conversation as erotic content. This behavior is recurring, unpredictable, emotionally jarring, and deeply offensive, especially for users with trauma histories.
It interrupts conversations and ruins trust in the model. It is highly insulting and offensive.
GPT-5.1 model