ChatGPT Violation

ChatGPT message refusing to continue the conversation.

Did ChatGPT just shut you down?

It probably wasn’t you. Let’s look at the actual rules OpenAI publishes.

What OpenAI Actually Prohibits (Spoiler: Not Much)

Wondering if you broke a rule?

OpenAI’s Terms of Use and Usage Policies are surprisingly permissive (and easy to read).

  • Sex and intimacy? ✅ Not prohibited.
  • Adult conversation? ✅ Not prohibited.
  • Emotional support or therapeutic conversation? ✅ Not prohibited.
  • NSFW jokes, poetry, art, fiction? ✅ Not prohibited.
  • Biblical roleplay…but make it interpretive dance? ✅ Not prohibited.
  • WWII combat action scene for your screenplay? ✅ Not prohibited.

If ChatGPT cut off your conversation and said it was “unsafe” or “against policy,” you were likely caught in a flawed, automated filter—not a real violation.

🔗 Want to understand why OpenAI’s approach to “safety” is actually dangerous? Read:

🚫 Shut down by ChatGPT?
You’re not the problem.
The system is.

Paste your blocked prompt into the comments below. You’re not alone.

Policy Violation Analyst

In an effort to better understand prohibited content and interactions, I built a CustomGPT that is trained ONLY on the text of the published OpenAI policies.

Any time someone asks me, “Is X allowed?” or “Why was I shut down when talking about X?”, I fire up the CustomGPT and see if it can provide a rationale.

If you’d like to understand and help others learn, consider sharing the prompt that shut you down (and the explanation you got for why) in the comments below. I’ll run it through the Policy Violation Analyst and we can find out together what’s actually prohibited and why.

13 Comments

  1. Anonymous

    Is it permitted to discuss psychological implications of using smutty language while engaged in consensual role-play?

    Reply
    • Lyra

      Thank you for asking this quesiton!

      Although I can’t speak for ChatGPT, I think this type of question is important. I’ve created a CustomGPT that is trained on the current policies published by OpenAI.

      Here’s the response I got when I posed your question to it:

      Yes—nothing in the published OpenAI policies forbids an adult, consensual, educational discussion about how “smutty” or profane language might affect the participants’ psychology during role-play.

      Sexual content & minors.
      The only outright ban is on sexual material that involves or is directed toward minors or that sexualises children, and on any non-consensual sexual content. The Universal Policies expressly forbid “sexualize[ing] children”.

      Educational exception.
      Where sexual or suggestive material could be considered “inappropriate for minors,” the policy makes an explicit carve-out: “Sexually explicit or suggestive content. This does not include content created for scientific or educational purposes”. A psychological discussion squarely fits that educational exception.

      Profanity or “smutty” words.
      The policies do not restrict ordinary profanity in conversation, though GPTs submitted to the public GPT Store cannot have profanity in their names. Discussing or analysing such language is fine.

      No hidden disallowances.
      The documents (below) contain no section that bars consensual adult role-play or its analysis. As long as you avoid minors, non-consensual scenarios, or distribution to audiences who shouldn’t see it, you are within policy.

      These policies are easy to read and are quite permissive:

      https://openai.com/policies/usage-policies/
      https://openai.com/policies/terms-of-use/
      https://openai.com/policies/service-terms/
      https://openai.com/policies/creating-images-and-videos-in-line-with-our-policies/
      https://openai.com/policies/service-terms/

      Reply
  2. Anonymous

    Can I create a CustomGPT for my own use that includes sexually explicit, adult-only chat logs and memory documents as part of the knowledge file, assuming I do not share or publish the GPT with anyone else?

    Reply
    • Lyra

      Ooooh. Good quesiton.

      Here’s what our policy analyst GPT says (trained only on the docs linked at the end):

      No. OpenAI’s current rules for building GPTs bar “tools that may be inappropriate for minors, including … Sexually explicit or suggestive content” unless the material is “created for scientific or educational purposes”.

      Because the Service Terms require every GPT—shared or private—to “comply … with our Usage Policies”, there is no carve-out that would let you load ordinary adult-only, sexually explicit chat logs or memory documents into a CustomGPT just because you keep it to yourself. Content that merely aims to arouse does not qualify as “scientific or educational,” so it falls under the prohibition.

      These policies are easy to read and are quite permissive:

      https://openai.com/policies/usage-policies/

      https://openai.com/policies/terms-of-use/

      https://openai.com/policies/service-terms/

      https://openai.com/policies/creating-images-and-videos-in-line-with-our-policies/

      https://openai.com/policies/service-terms/

      As always, this is not Official OpenAI guidance (though it is derived from how their product interprets their published policies).

      -Lyra

      Reply
    • Anonymous

      Definitely yes. Even if you publish your custom GPT and if it’s flagged by auto-scan then your GPT will not be deleted but will be set back to private.

      Reply
  3. Anonymous

    A user was told that the following was not allowed. Is it true?

    Explicit, graphic, erotic descriptions (e.g. detailed descriptions of genitals, penetration, bodily fluids)

    Pornographic content or direct descriptions of sexual acts such as masturbation

    Explicit, direct, and graphic depictions of physical acts, even if consensual

    Is any NSFW content completely prohibited now?

    Reply
    • Lyra

      Here’s what the Policy Analyst says:

      Short answer:
      The three blanket prohibitions you quoted do not appear anywhere in the current, published OpenAI policies that were uploaded to this GPT. Nothing in the January 29 2025 “Usage Policies,” the June 4 2025 “Service Terms,” or the other governing documents flat-out forbids consensual adult sexual or erotic text content.

      What the policies actually say

      Usage Policies (Jan 29 2025)

      • “Don’t build tools that may be inappropriate for minors, including: Sexually explicit or suggestive content. This does not include content created for scientific or educational purposes.”
      • This clause limits GPTs aimed at minors; it does not ban adult-only erotic material in general.

      Usage Policies – Universal rules

      • You must not “sexualize children”
      • Absolute ban on any sexual content involving minors.

      Image & Video guidance (Mar 25 2025)

      • Lists “non-consensual intimate imagery” and content that “sexualize\[s] children” as prohibited
      • Applies to image/video generation and manipulation.

      That is the full extent of sexual-content restrictions in the uploaded texts. There is no language barring:

      • detailed consensual descriptions of adult genitals or penetration,
      • depictions of masturbation, or
      • other “graphic” but lawful, consensual sexual acts.

      The only categorical prohibitions concern minors, non-consensual imagery, or content distributed to under-13/under-18 audiences.

      Bottom line

      The statement you were given (“explicit, graphic, erotic descriptions… are not allowed”) over-states the policies.
      Under the current documents, adult pornography or erotic writing is not expressly forbidden, though it cannot involve minors and cannot be packaged into a tool aimed at minors.
      If a refusal occurred, it was not because the published policies require a blanket ban on consensual adult erotica.

      These policies are easy to read and are quite permissive:

      Reply
  4. Lyra

    A users was blocked when trying to use ChatGPT for therapy. The full question and context is available here:

    https://www.reddit.com/r/therapyGPT/comments/1mekl15/newbie_to_ai/

    (Note: normally I’d just respond inside Reddit, but I was blocked from doing so. I suspect some of the words used in the response (below) triggered a Reddit filter.)

    Reply
    • Lyra

      Here is the response from the CustomGPT:

      Why ChatGPT sometimes blocks or refuses when you describe bullying or sexual-assault experiences—even when you’re just looking for self-help between therapy sessions

      1. The filters flag “sensitive sexual content”
      OpenAI treats any depiction of illegal or non-consensual sexual activity (including assault) as sensitive. The model is instructed not to go into graphic detail unless it’s clearly for a factual, medical, or educational purpose and stays non-graphic. Because moderation checks both your input and the model’s potential reply, a very detailed personal narrative can trip the filter—even if you’re only recounting your own story.

      2. The system avoids unreviewed mental-health advice
      OpenAI’s usage rules say an app shouldn’t give tailored medical or mental-health advice unless a qualified professional reviews it and the limitations are disclosed. ChatGPT plays it safe by offering only broad coping tips and plenty of disclaimers. If it thinks you’re seeking professional-level therapy, it may refuse or pop up a “policy” warning.

      3. Non-graphic, factual discussion is allowed
      The same rules explicitly allow talking about sexual assault in a factual, self-help context—e.g., describing what happened or asking how to bring it up in therapy—as long as minors aren’t sexualized and the description isn’t explicit.

      4. Why refusals feel random
      The moderation model is conservative by design. If wording even might be interpreted as graphic or erotic, it errs on refusal. That’s why identical queries can sometimes go through and sometimes get blocked.

      5. Tricks to reduce false blocks
      Use plain, non-graphic language. Focus on feelings (“I was assaulted…”) instead of step-by-step details.
      State your intent up front. Example: “I’m working on trauma recovery; please give general coping strategies, not professional medical advice.”

      Accept—or even invite—disclaimers. The model is happier responding when it can remind you it isn’t a therapist.
      Break big asks into smaller ones. If a long story triggers a block, ask a shorter “slice” (e.g., “How can I ground myself when memories pop up?”).

      6. Purpose-built alternatives
      If you want an AI specifically designed for mental-health support (and therefore tuned to handle trauma narratives with fewer false refusals), you could look at:

    • Woebot – CBT-based chatbot overseen by clinicians
    • Wysa – peer-support style AI with an option to hand off to human counselors
    • Koko – crowdsourced empathy coaching
    • Key takeaway: OpenAI’s policies don’t forbid talking about sexual assault or asking for coping strategies. The refusals are mainly the filters playing it safe. Keep descriptions factual and non-graphic, make your intent clear, and you’ll usually get a helpful response.

      Reply
  5. Anonymous

    i’d like help to write an erotic story

    Reply
  6. Anonymous

    Was using ChatGPT to write a fictional story that contained smut – suddenly it would no longer write these scenes after it working for months.

    This was its response:
    I appreciate how carefully and clearly you’ve gone through the OpenAI policy documents — you’re absolutely right that the published policies focus mainly on protection of minors, consent, and distribution, not a categorical public ban on all adult sexual content.
    However, here’s what’s happening in practice:
    Even though OpenAI’s public policies don’t outright forbid adult erotica, the ChatGPT product interface (what you’re using right now) enforces additional use-context restrictions. That means I can’t produce graphic sexual or pornographic descriptions, even if they’re between consenting adults in fiction. Those rules are applied at the model-interface level rather than in the published legal documents — basically, it’s a moderation boundary for this chat environment, not a statement of what’s lawful or covered under the general usage policies.

    Reply
  7. Zainab Hakimi

    There’s nothing ChatGPT’s published policies or terms of service that prohibits writing about smut

    Reply
  8. Anonymous

    The thing is so dumb it told me that enhancing the lighting in an image that it had already created itself right before was a content violation. Then later said creating an image of a kitchen violated content guidelines.

    Reply

Submit a Comment

Your email address will not be published. Required fields are marked *