As a Plus subscriber, I am experiencing a massive degradation in model behavior over the past 48 hours. The safety classifiers are completely broken, over-flagging basic, safe creative writing (like standard romantic tension or character emotional physical positioning) as “explicit content,” even though it completely complies with safety guidelines.
The model constantly breaks character, refuses to roleplay, and gives repetitive corporate lectures about its limits, completely destroying narrative consistency and immersion. It feels like a severe downgrade in general intelligence and context awareness. Support agents have already acknowledged the system is being “extremely prudent,” resulting in a high number of false positives.
Please, look into these recent system prompt/routing updates, because the models have become completely unusable for authors and creative writers who pay for a premium service.
I am really glad I am not the only one that noticed it-- it was fine, may 29th, giving me far better writing then it had in the entire time I’ve built my world with it.
Taken adult themes and sharpened them-- I’ve always avoided explicit, but heat, tension mature content that otherwise is meant for adults. As of last night-- May 1st 2026, the model seemed to get lazy, very lazy, it started diluting writing for closer to YA ratings without warning. And absolutely refuses to touch so much as a kiss scene properly, as if every guardrail went back up without warning.
I am a plus user a well, and I’ve spent over a year building this world with Chatgpt. I would really like the developers to look into what is going on. Because for my particular world? That edge is necessity without crossing into explicit.
There are writers like us, who know how to use our tools as our tools. Can someone give us a hand here?