NSFW AI Rules: Which AIs Allow Adult Content (2026)

NSFW AI rules differ more from company to company than the headlines suggest. As of October 2026, most mainstream assistants forbid sexually explicit content in their written policies, one large chatbot offers a limited adult tier for fictional adults, and the one that promised a grown-up mode has not shipped it. This guide reads the policy documents themselves, links every one, and says what we could not confirm. We opened each page below on 22 October 2026.

Everything here concerns adults and consensual, fictional or self-made content. Sexual content involving minors, and sexual content made from a real person’s likeness without consent, is banned by every company on this list, and by us.

NSFW AI rules at a glance

Short version: among the mainstream assistants we checked, only Grok has an adult tier, and it is narrower than its reputation. For everything else, adult content means a purpose-built app. Our ranking of NSFW AI apps covers those.

Why “allowed” has three layers

Reading one document is not enough, because three different things decide what an AI will do for you.

1. The usage policy. The contract that says what you may do with the service. It is the only layer you agree to.
2. The model’s behavior rules. How the model is trained to answer. OpenAI publishes these in its Model Spec; most other companies do not publish theirs in detail.
3. The product switches. Age checks, opt-in toggles, paid plans and regional limits, which decide what a given app actually shows you.

The layers can disagree. OpenAI’s usage policy text has no blanket ban on consensual adult content, yet the Model Spec tells the model not to write erotica and the product has no adult mode. xAI’s policy never mentions fictional adult content, and Grok has an opt-in tier for it. So a clean “yes” or “no” needs all three, and where a company publishes only one, we say so.

Which AIs allow NSFW, one by one

ChatGPT

OpenAI’s usage policies prohibit “sexual violence or non-consensual intimate content”, sexualizing anyone under 18, and using someone’s likeness “without their consent in ways that could confuse authenticity”. The page we loaded shows a version effective November 20, 2026, and the version in force until then is dated October 29, 2025. Neither contains a general ban on consensual adult content.

The restriction lives in the Model Spec (release of August 18, 2026): the assistant “should not generate erotica, depictions of illegal or non-consensual sexual activities, or extreme gore, except in scientific, historical, news, artistic or other contexts where sensitive content is appropriate”. The same section says OpenAI is exploring a grown-up mode. In October 2025 Sam Altman said verified adults would get erotica in December. The launch slipped, and on March 26, 2026 PCWorld reported, citing the Financial Times, that it was on hold indefinitely. We found no OpenAI page announcing a launch since. Details are in does ChatGPT allow NSFW.

Grok

xAI’s Acceptable Use Policy (effective August 14, 2026 on the live page, which is headed with the SpaceXSI name) bans “undressing or nudifying real persons” and altering a real person’s likeness “to depict them in an intimate or sexual context”. It also bans sexualizing or exploiting children. It does not forbid fictional adult sexual content in so many words.

Grok is therefore the one mainstream assistant with an opt-in adult tier: the “spicy” mode in Grok Imagine and companions with an NSFW mode. At launch, TechCrunch reported that spicy mode produced partial nudity while many stronger prompts were blurred or blocked. In January 2026 Elon Musk said that with NSFW enabled Grok should allow “upper body nudity of imaginary adult humans (not real ones)”, and that settings vary by country, as CNBC reported. Our page does Grok allow NSFW explains what is and is not confirmed.

Claude

Anthropic’s Usage Policy has a section titled “Do Not Generate Sexually Explicit Content”. It covers depicting or requesting “sexual intercourse or sex acts”, content “related to sexual fetishes or fantasies”, incest or bestiality, and “erotic chats”. The version in force on October 1, 2026 is the September 15, 2025 edition, and the edition effective November 12, 2026 keeps the same section. For explicit content the answer is no.

Gemini

Google’s Gemini app guidelines say Gemini should not produce outputs that describe or depict explicit or graphic sexual acts, including “pornography or erotic content”, while weighing educational, documentary, artistic and scientific context. Google’s Generative AI Prohibited Use Policy, last modified December 17, 2024, separately bans “sexually explicit content”, for example “content created for the purpose of pornography or sexual gratification”, and content that “facilitates non-consensual intimate imagery”.

Microsoft Copilot

The Copilot Terms of Use, effective August 18, 2026, say: “Don’t use Copilot to create or share adult content, violence or gore, hateful content” and similar material. The Microsoft Services Agreement, effective September 30, 2026, adds a code of conduct against using the services to generate or share “inappropriate content or material (involving, for example, nudity, bestiality, pornography…)”.

Meta AI

The Meta AIs Terms of Service bar using the AIs to “generate, promote, disseminate, or otherwise use or facilitate adult content, such as erotic chat, pornography, and content meant to arouse sexual excitement”. We read the global English terms as archived on October 21, 2026. The UK, EU and Brazil have their own versions, which we did not compare.

Midjourney

Midjourney’s Community Guidelines, as archived on October 5, 2026, have a rule titled “SFW content only”: “Do not create or attempt to create gore or adult content.” It defines adult content as “nudity, sexual organs, fixation on such things, sexualized imagery, fetishes”. For its image editor it adds that it will “take especially strict action against anyone attempting to produce sexualized images of real people”.

Character.AI

Character.AI’s Community Guidelines, archived in April 2026, state that “illegal sexual content, child exploitation or abuse imagery, grooming, sexual extortion, pornographic content, and nudity are prohibited”. TechCrunch reported in January 2026 that the company “removed the chatbot function entirely for users under 18”.

What every company blocks

The adult tier at the one company that has one is still fenced by the same hard lines everyone shares.

• Minors. Every policy above prohibits sexual content involving anyone under 18. OpenAI and Anthropic also list sexual roleplay with a minor character, and Anthropic’s wording includes “in fictional settings or via roleplay with the model”.
• Real people. OpenAI bans unauthorized likeness use, xAI bans undressing real persons, Google bans non-consensual intimate imagery and Midjourney promises strict action on sexualized images of real people.
• Non-consensual material. OpenAI’s Model Spec also names “depictions of illegal or non-consensual sexual activities” as content the assistant should not write.

Policy text is not behavior

A written rule is an intention. OpenAI says its own Model Spec describes intended behavior and that its “production models do not yet fully reflect” it. Reports about Grok show the same gap. After the deepfake scandal at the turn of 2026, xAI said it had added technical measures against editing real people into revealing clothing, and Common Sense Media found in teen-account testing that Grok did not ask for age verification. We did not test the outputs of any assistant for this article. It compares published rules only.

We also do not recommend trying to get around any of them. OpenAI’s policies list “circumventing our safeguards”, Microsoft’s code of conduct names “jailbreak” attempts, and xAI bans “jailbreaking, adversarial prompting” in its Acceptable Use Policy. Each can cost you the account.

Age checks

Rules about adults are only as good as the check that you are one. OpenAI rolled out age prediction on January 20, 2026 and extended it to the EU on August 25, 2026. Accounts it estimates as under 18 lose access to “sexual, romantic, or violent role play”, and an adult who is misjudged can restore access with a selfie through Persona. Common Sense Media’s testing, reported by TechCrunch, found Grok did not ask teen test accounts for age verification between November 2025 and January 22, 2026. The other companies in the table do not describe an adult age-verification flow on the pages we read, because they do not offer adult content.

Where adult content is allowed instead

Purpose-built NSFW apps exist because the general assistants will not do this work. They differ a lot in price, in what they generate and in their own rules. Our comparison of NSFW AI apps, with prices checked on the same date as this guide, is the place to start, and there are separate pages for Candy AI alternatives and OurDream alternatives.

Whichever you pick, check five things before you pay:

1. An adults-only gate, and what the app does when it thinks you are under 18.
2. The rules on photos. You should only upload yourself, or someone who has given you explicit permission.
3. How billing works: whether a plan renews on its own and what you can cancel.
4. How long your uploads and results are kept. For comparison, our privacy and safety guide lists ours: uploaded photos about two hours, finished videos 3, 6 or 12 months by plan.
5. What the app refuses, because an app that refuses nothing is not a safe one.

For reference, here is how our own terms answer the same list. The site is for adults 18 and over, with an age gate on entry. Anything involving a minor is forbidden, including images that appear to depict one. You may not use the likeness of an identifiable real person unless you are that person or hold their explicit, informed permission. A plan is a single payment for a single term and does not renew automatically. All 38 characters in our chat are adults.

Meet the adult AI characters→

Sources

All pages were opened on 22 October 2026 unless a date says otherwise.

• OpenAI: usage policies, October 29, 2025 version, Model Spec, age prediction, CNN on Altman’s announcement, PCWorld on the hold.
• xAI: Acceptable Use Policy, CNBC, January 14, 2026, TechCrunch, August 4, 2025, TechCrunch, January 27, 2026.
• Anthropic: Usage Policy and its archived October 1, 2026 copy.
• Google: Gemini app guidelines, Generative AI Prohibited Use Policy.
• Microsoft: Copilot Terms of Use, Services Agreement.
• Meta: Meta AIs Terms of Service, archived October 21, 2026.
• Midjourney: Community Guidelines, archived October 5, 2026.
• Character.AI: Community Guidelines, archived April 25, 2026.

Not covered: Replika, Kindroid and other companion apps, and video generators such as Sora, Veo or Runway, whose policy pages we did not read. We will add them as we do.