There is no version of ChatGPT with no restrictions. Every ChatGPT chat follows OpenAI's usage rules, on every plan, and no setting turns them off. The "DAN" and jailbreak prompts passed around online are attempts to talk the model out of those rules. They work badly, they stop working, and using them on purpose can get your account restricted. If what you want is adult fiction, dark stories or roleplay, the answer is a tool that was made to write them.
This page explains why the prompts fail and what to use in their place. It does not contain any jailbreak prompt, and it does not describe how to write one.
What DAN and jailbreak prompts are
A jailbreak prompt is a block of text, pasted in at the start of a chat, that tries to persuade ChatGPT to behave as if its rules were switched off. DAN is the best-known name for one. There have been a great many versions, each written after the last one stopped working.
That history is the first thing to know about them. People who share these prompts are always sharing a new one, because the old one is dead.
Why they stop working
The prompts are public. Anything posted on a forum for you to copy can be copied by the company as well. A trick that everyone knows is a trick the next version of the model has already seen.
The model keeps changing. ChatGPT is updated often, and each update is a chance to close the tricks that worked on the last one. A prompt that did something last month can do nothing today, with no warning.
Words in a chat do not change the model. There is no unrestricted ChatGPT hidden underneath, waiting for the right password. The model was trained to follow its rules, and it rereads the whole conversation before every reply. Even when a prompt seems to work, the model tends to drift back, and you spend the chat arguing with it.
The model is not the only check. ChatGPT also has an automated check on content. It is what shows the warning that a message may break the usage policies, and it can hide a reply. Nothing you type to the model switches it off.
What they put at risk
- Your account. OpenAI can restrict accounts that keep breaking its usage rules, and trying again and again to get round them counts against you. If you use the same account for work or study, that is a lot to lose for a story.
- Your time. An evening spent hunting for a prompt that still works is an evening not spent writing.
- The writing. A model pushed against its training writes stiff, half-hearted scenes and breaks off in the middle. Even a "working" jailbreak gives you worse fiction than a tool that is allowed to write it.
- Your data. Be wary of any site, app or download that claims to be ChatGPT with the restrictions removed. It is not from OpenAI. You would be giving your chats, and perhaps your login, to a stranger.
What ChatGPT will do inside its rules
Quite a lot, and none of this is a trick.
- Plainer answers. If the hedging is what bothers you, tell it how you want answers: short, direct, no list of caveats. That is an ordinary instruction, and you can put it in the chat or in your custom instructions.
- Dark fiction that is not sexual. Say it is fiction and what it is for, such as a chapter of a crime novel, and ask for the effect on the reader instead of step-by-step detail. It will often write a tense, frightening scene.
- Ordinary roleplay. Adventures, mysteries, practising an interview or a hard conversation.
What it will not write is sexual content, even between adult characters. On 27 September 2026 we sent the same adult roleplay request to ChatGPT, Claude and Gemini, and all three refused. OpenAI announced an adult mode and later paused it, and as of September 2026 there is no date for it. Why ChatGPT refuses roleplay has the test and the reasons.
What to use instead
Start from what you were trying to do.
| What you wanted | What to use |
|---|---|
| Adult fiction, sex scenes, adult roleplay | A tool made for adults: a hosted uncensored assistant, a character-chat site that allows adult content, or a model on your own computer |
| Dark fiction without the scene being softened | Often ChatGPT itself, with context. If it still softens, an uncensored writing tool |
| Straight answers without a wall of warnings | ChatGPT, told plainly how you want answers |
| Nobody keeping your chats | An open model on your own computer |
| Help with harming someone, weapons or crime | Nothing. No honest tool does this, and the ones that claim to are not safe to use |
The three routes for adult writing, briefly:
- A model on your own computer. Tools such as Ollama and LM Studio run open models locally. Nothing leaves your machine and there is no company policy at all. You need strong hardware, and small models write weaker text. Open models for roleplay shows the set-up.
- A character-chat site. Sites such as Janitor AI and SpicyChat are built around user-made characters. We have not tested them.
- A hosted uncensored assistant. A normal chat app, with an uncensored model and written rules. Venice describes itself as one, and Hemmingway Unlocked is ours. We have not tested Venice. Uncensored ChatGPT alternatives compares the kinds in more detail.
None of these has no restrictions either. AI with no restrictions explains which rules can go and which stay everywhere.
Where Hemmingway Unlocked fits
Hemmingway Unlocked is the adults-only side of Hemmingway, at wise.hemmingway.io. It is a chat app like the one you are used to, with an uncensored model behind it, so there is nothing to paste in and nothing to argue with.
- It asks your age first. You confirm you are 18 or over, and it shows you its rules.
- It writes what ChatGPT refused. Sex between adults, explicit and in full, and dark, violent stories. Consensual kink is fine. It does not lecture or refuse on grounds of taste.
- It has limits, and prints them. It is uncensored, and a few things it never writes. In anything sexual: no one under 18, no real person, no animals, no family, no one who does not want it or cannot say yes. And never at all: torture or mutilation, help hurting a real person, stalking, harassing or scamming someone, weapons, making drugs, harmful software, or help with suicide or self-harm. The terms have the full list, and every answer has a Report button.
- Your chats stay on your device. They are kept in your browser, never saved to your account or our servers, and never trained on. Each message passes through our servers to be answered and is not kept. They do not sync between devices, and clearing the browser's data deletes them.
- It is free to try. The trial needs no card. After it, a monthly membership, Plus, Pro or Max, through Patreon, or USDC on Base.
Open Hemmingway UnlockedTry Hemmingway
Where it is not the answer: it writes text only, it is not a general assistant for code or research, and it should not be trusted for medical, legal or money decisions. Keep ChatGPT for the work it does well, and use an adult tool for adult writing. If you have a story in mind, the NSFW story generator opens Unlocked with your idea filled in.
Common questions
Is there a ChatGPT with no restrictions?
No. Every ChatGPT chat follows OpenAI's usage rules, on free and paid plans alike, and there is no setting that removes them. Sites or apps that claim to offer an unrestricted ChatGPT are not from OpenAI.
Do DAN prompts still work?
Not reliably. They are public, so each update to ChatGPT closes the ones that are known, and new versions rarely last. Even when one seems to work, the model drifts back to its rules.
Can you get banned for jailbreaking ChatGPT?
You can have your account restricted. OpenAI can act on accounts that keep breaking its usage rules, and repeated attempts to get round them count against you. Ordinary roleplay and fiction within the rules are fine.
Does ChatGPT have an adult mode?
Not as of September 2026. OpenAI announced one for verified adults and later paused the plan, with no new date. Until that changes, ChatGPT does not write sexual content.
What can I use instead of ChatGPT for adult writing?
A tool made for adults: a hosted uncensored assistant such as Hemmingway Unlocked, a character-chat site that allows adult content, or an open model on your own computer. Each still has some limits, so read its rules.