No app on this page defines “uncensored” in the rules we read. CrushOn’s site menu says “Unfiltered”, yet its Community Guidelines, read on 30 October 2026, still say: “Do not create or depict characters who are minors or whose age is not clearly defined as over 18.” Candy AI’s terms, revised 22 September 2026, go further and say its moderation “may moderate or alter your request before an output is generated”.
We read the terms, guidelines and safety pages of nine apps on 30 October 2026, and re-read all of them on 31 October 2026. The pages are listed in the sources at the end. We have not paid for these apps and did not test what any filter lets through, so nothing here carries a score. The order follows how much each app publishes about its limits, most first, and commission plays no part in it. How we test explains what a paid test would add. Every app on this page is aimed at adults.
Candy AI and OurDream pay us a commission if you subscribe through the links on this page. The other seven apps earn us nothing from this page.
Limits each app publishes, side by side
“Unfiltered” is a marketing word, and the rules behind it differ in three ways: what the app blocks, whether the rule reaches your private chats, and how often a human reads what was flagged. The table covers the first two.
| App | How its own pages describe adult content | What its rules say it blocks | Reaches private chats? | Page date |
|---|---|---|---|---|
| OurDream | “Most of what people make on it is adult creative work” | Minors, real people, terrorism, excessive gore, self-harm | Yes, stated outright | Hard-limits page modified 1 Oct 2026 |
| CrushOn | “Unfiltered” in the site menu | Characters who are minors or of unclear age, realistic images of real people, graphic violence | Interactions are monitored | Terms 22 Sep 2026, guidelines undated |
| SpicyChat | Explicit chatbots “must be marked as NSFW” | Characters under 18, extreme violence mixed with sexual content, photorealistic images mistakable for real people | Rules cover “Chats and/or in Avatars”; terms say operators “may read all messages” | Guidelines 21 Oct 2025, terms 23 Jul 2026 |
| Candy AI | No label found in the terms | “Prohibited content”, zero tolerance for child abuse material; guidelines ban images resembling real people | Flagged content can be read by staff or partners | Revised 22 Sep 2026 |
| HammerAI | Terms say the services hold adult material | “Prohibited Content”, characters portrayed as underage, characters of real people | Terms say chats must not contain Prohibited Content, though HammerAI says it cannot access them | Terms 29 May 2026 |
| Nastia | “Uncensored” on the sign-up button | Sexualising minors, trafficking, deepfakes, prompts naming public figures | Private service; uploads disabled | Undated |
| Janitor AI | NSFW off by default on mobile, on by web setting | Explicit avatar images, sexualising minors including in fiction | Outputs can be “filtered, moderated, or refused” | Terms 8 Oct 2026 |
| NovelAI | No label found | No filter statement found in its FAQ or terms | Stored encrypted; terms say you hold the key | Terms 8 Dec 2025 |
| SillyTavern | Interface only | Depends on the model you connect | Runs on your own machine | Docs read 29 Oct 2026 |
“Not found” means the pages we read do not mention it. Three well-known apps are not in the table: see the section on apps outside the table.
What the rules say about minors and real people
Seven apps publish a rule about characters who are minors: Candy AI, CrushOn, HammerAI, Janitor AI, Nastia, OurDream and SpicyChat. Janitor AI’s terms ban the sexualisation of minors “in any form, including fictional depictions”. OurDream’s version is the widest: it judges age “on the whole picture, not a stated number”, so a stated age of 18 does not rescue a character framed as a child. SpicyChat requires that “All Chatbots created on the platform must be portrayed as adults over 18 years old.” HammerAI’s content guidelines, last updated 7 April 2026, say characters “where the user can be portrayed as underage will be removed”.
Six apps limit depictions of real people. CrushOn’s guidelines say not to post “real images or overly realistic AI-generated images of any individuals”. OurDream says “We do not host real-person likeness.” SpicyChat prohibits photorealistic images “that could reasonably be mistaken for real human beings”. Nastia says prompts naming public figures “are refused” and bans “non-consensual and deceptive deepfakes of real people”. Candy AI’s community guidelines ban “images that resemble real persons”, and HammerAI’s content guidelines say it does not allow characters that use images of real people or represent real people.
If an app you are considering is silent on both topics, read its terms before you pay, because silence does not show that nothing is blocked.
Do these limits reach your private chats?
OurDream says yes, in writing: “The hard limits below apply everywhere on the platform: public submissions, private chats, and private generations.” Its moderation page adds that “Every prompt is classified before any model runs against it”. The same hard-limits page says the private bar is looser than the public one for a few categories, such as parody and specific fetishes. That sits awkwardly with the homepage claim that chats use “end-to-end encryption” and that no one can read them.
Candy AI is the clearest case of a filter that acts before the reply. Section 11 of its terms says that “our content moderation controls may moderate or alter your request before an output is generated”. Section 5.4 adds that conversations are “generally private”, but if its controls flag something, “we or authorized partners may access and manually review the flagged content”.
CrushOn says it uses “a combination of human moderators and automated systems to monitor and review Crushon accounts, characters, and interactions”. SpicyChat’s terms say operators “may read all messages” you send. HammerAI concedes the limits of its own filters: “While HammerAI employs safety filters to reduce the likelihood of harmful outputs, no filtering system is perfect.” Its terms also say HammerAI “cannot access your conversations”, yet that you agree your conversation will not contain prohibited content.
Our AI girlfriend app privacy page compares how long several of these apps keep chat data, which matters more once you know staff can read flagged messages.
The nine apps, from most published limits to fewest
OurDream: the most detailed rulebook, and it covers private chats
OurDream’s safety hub lists “hard limits” with “no creative-framing exceptions” and says moderation runs as “defense in depth”: classifiers on input and output, behaviour signals, human reviewers and community reports. It promises “We tell you when your content is removed, and why.” OurDream’s help centre listed plans at $9.99 a month billed yearly or $19.99 monthly on 29 October 2026, and our OurDream AI review has the rest. Skip it if you want to roleplay without an input classifier reading each prompt. Visit OurDream (affiliate link: we earn a commission if you subscribe).
More on OurDream: review · is it safe? · promo codes · alternatives
CrushOn: “Unfiltered” in the menu, a full rulebook behind it
CrushOn’s Terms of Use, last modified 22 September 2026, tell users not to submit material that is “NSFW, pornographic, or otherwise sexually explicit” as a User Submission, and say accounts can be ended for violating “our policies on adult nudity and sexual activity”. We read that clause as aimed at public submissions, but the terms do not say whether private chats are exempt. Its guidelines ask users to “Avoid depicting explicit sexual activity and stimulation” in imagery, and ban graphic violence. Our CrushOn AI overview covers its pricing and refunds. Skip it if a clear private-chat rule matters to you, because the pages we read do not state one.
SpicyChat: adult content is a label creators must apply
SpicyChat’s guidelines say “Chatbots with explicit language or imagery must be marked as NSFW”, so the platform sorts adult content by label. Its unacceptable-content list applies “in Chats and/or in Avatars”, and it states that “Mixing extreme violence with sexual content is strictly prohibited.” The guidelines page was last updated on 21 October 2025. SpicyChat is also covered in our roleplay chatbot roundup, which compares its lorebook and memory limits. Skip it if you want a recent policy: its guidelines date from October 2025, and its terms let operators read your messages.
Candy AI: the terms say your request can be altered before a reply
Besides the pre-output clause quoted above, Candy AI’s terms say users “will not be refunded or reimbursed for moderation actions taken in response to violation of our Policies”. It states a “zero-tolerance policy toward Child Sexual Abuse Material” and reports known cases to authorities. Read our Candy AI review for its plans. Skip it if you want to know in advance what counts as a prohibited request, because the terms point to separate policies. Visit Candy AI (affiliate link: we earn a commission if you subscribe).
More on Candy AI: review · is it safe? · how to cancel · alternatives
HammerAI: filters on, with an honest disclaimer
HammerAI’s terms say its services contain “descriptions of nudity, and/or sexual activity” and are for adults only. They also warn that AI output “may contain inaccuracies, biases, or inappropriate content despite our filtering efforts”. Its FAQ says local models on the desktop app run on your own device, and that cloud messages are sent to its servers but “are not stored or used for training”. Its privacy policy is less generous: it says third-party model providers may store chat content and use it for AI training. Our HammerAI overview covers the free plan. Skip it if you want one policy page: its rules sit across the terms, content guidelines and compliance pages.
Nastia: “uncensored” on the button, hard limits one click away
Nastia’s compliance page invites you to “Create my uncensored AI companion” and does not define the word. The same page describes a “zero-tolerance policy for content that sexualizes minors”, refuses prompts naming public figures, and says “Nastia is a closed, private service” with image and video uploads disabled. It lists no date. Our Nastia AI overview has the pricing it publishes. Skip it if you need a dated policy.
Janitor AI: adult content is a setting, not a default
Janitor AI’s help centre says: “NSFW content is disabled by default on the mobile application.” A web setting turns it on, and avatars may not carry “Explicit nudity, pornographic content, or overtly sexual imagery”. Its terms show that outputs can be “filtered, moderated, or refused”, and that Router credits can still be charged when they are. We re-read these pages on 31 October 2026. See our Janitor AI alternatives page. Skip it if you want adult content on mobile, where it is off by default.
NovelAI: no content-filter statement in the pages we read
NovelAI’s terms (effective 8 December 2025) require users to be 18 or older. Its FAQ and terms say nothing we could find about content filtering, in either direction. The FAQ says only that image reference features are kept out of its Explore gallery because they are “harder to moderate and easier to misuse”. A missing statement says nothing about whether a filter exists. Skip it if you need a written content policy before subscribing.
SillyTavern: no policy, because the model decides
SillyTavern is software you install, and its docs describe it as “only an interface”, so you need “an LLM backend to provide inference”. It has no content policy of its own to read. Whatever filtering exists comes from the model provider you connect, and that provider’s terms apply. Skip it if you want a hosted app that works the moment you sign up.
Apps outside the table
Three apps people often call unfiltered are not in the table. Character.AI’s guidelines still returned a Cloudflare challenge on 31 October 2026, and we did not work around it. Kindroid’s legal page, updated 30 October 2026, is now readable: its terms ban using a person’s “name, likeness or persona” without permission, and point to separate moderation guidelines we have not read. Nomi’s terms, effective 16 January 2026, require users to be 18 or older, and we found no filter, moderation or private-chat clause in them. Our Kindroid overview and Nomi overview cover those apps’ pricing and memory, and we will add both apps to the table once we have read their full moderation rules.
Which app’s rules fit what you need to know first
- To see exactly which categories are blocked: OurDream, whose safety hub lists them one by one.
- To see how adult content is sorted: SpicyChat, where creators label it and the guidelines name the limits.
- To run a model with no policy of your own: SillyTavern, with the model provider’s terms as the real rules.
- To see a filter’s limits admitted in writing: HammerAI, whose terms say “no filtering system is perfect”.
Before you pay any of them, open the app’s terms and search the page for “moderate”, “review” and “private”. Those three words answer whether your prompts are altered, whether staff can read flagged chats, and whether a rule reaches private use. If you only want to compare free allowances, our free NSFW AI chat apps page does that on the same basis.
What changed on 31 October 2026
We re-read every app’s rules. We had missed three published rules: Janitor AI bans sexualised minors in fiction, Candy AI’s community guidelines ban images that resemble real people, and HammerAI bans characters of real people. The counts now read seven apps on minors and six on real people. SpicyChat’s guidelines carry a date (21 October 2025), and its terms let operators read messages. Kindroid’s and Nomi’s rules are now readable and are summarised below the table, and we added the private-chat lines from HammerAI, NovelAI and OurDream.