Independent dating app reviews 24 files open · facts last checked 22 NOV 2026
File No. 015

AI companion · JanitorAI Inc. (Delaware); payments via Dulova Investments Limited (Cyprus)

Janitor AI: the fact file

A free role-play chat site built on user-made characters, run by a Delaware company that can bill through a Cyprus affiliate. Adult role-play is allowed and its written rules on minors are detailed, but it checks ages only in Australia, Brazil and the UK, a 2025 research report named it among four platforms where it found sexualized minor personas, and its privacy policy says nothing about AI training or the outside companies that receive many users' chats.

Facts checked November 20, 2026 · jani.com ↗

Fact fileNo score yet

Good to know

  • The terms name the operator: JanitorAI Inc., a Delaware corporation according to its US trademark record. Charges can come from an affiliate in Cyprus, Dulova Investments Limited.
  • The free tier is the main product: unlimited chats with its own model, according to the company. The only plan, janitor+, costs $12.99 a month or $99 a year.
  • Its rules on minors are detailed. An adult age must be written into a bot's definition, and schooling typical of minors or "teenager" framing counts against a stated adult age.
  • Users in Australia, Brazil and the UK must pass a face, ID or AgeKey check run by k-ID. The company says it receives only a yes or no.
  • You can cancel janitor+ through Stripe's billing page and keep it to the end of the period. A deleted account can be restored for 30 days, according to the site's interface.
  • Its crisis protocol says a notice with crisis resources appears when a user writes about suicide or self-harm.

Watch out for

  • Everywhere else, the age check is a checkbox. The home page sent to a signed-out visitor on November 20, 2026 had no age screen, and 20 of its 34 listed bots carried the Limitless tag for sexual or romantic themes.
  • Graphika's March 2025 study found over 100 sexualized minor personas or scenarios across four platforms, JanitorAI among them. It gave no count for each. On November 20, 2026, the home page listed a Limitless role-play bot set at an anime's high school, with characters "aged to 18."
  • The privacy policy doesn't say whether your chats train its own model, and doesn't name the outside AI companies your chats reach through a proxy or its Router.
  • Fees are generally non-refundable. Router credits are final, can expire after 12 months, and are canceled if you delete your account.
  • Its own documents disagree. The terms ban content that "depicts" self-harm or sexual violence; the help center allows both as tagged story themes. Pages also differ on the Router launch bonus and on account deletion.
  • Disputes go to individual arbitration under Delaware law, and its liability is capped at the greater of 12 months' fees or $50.

Janitor AI is a free website and app for role-play chats with characters that its users create.12 It allows sexual role-play between adult characters.34 A March 2025 company post said it “literally exists because of other sites strict filters.”5 It names its operator, and its written rules on minors are detailed. But it checks ages only in three countries, a 2025 research report named it among four platforms where it found sexualized minor personas, and its privacy policy says nothing about AI training or the outside AI companies that many users’ chats go to.678

What it is

The terms describe a platform for creating characters and chatting with them, and with other users’ characters, through jani.com and the official apps.1 Creators rate each character “Limited” or “Limitless.” Limited means no sexual or romantic themes. Limitless is required for “Sexual or romantic themes,” “NSFW plotlines” or explicit dialogue, and doesn’t make banned content allowed.49 Character images may not show nudity or sexual activity, so the explicit part is text.310

Replies come from one of three places:

  • JLLM, the company’s own model. The help center calls it “the default model,” free and built in.11 An April 20, 2026 changelog entry says its base model moved to a new architecture “heavily tuned with high quality gemini and opus data.”12
  • Your own proxy. You can paste in an API key from an outside AI provider. The terms allow “configuring an API proxy in your settings,” and the help center walks users through connecting OpenRouter and a DeepSeek model.113 Character pages carry a “proxy allowed” label.14
  • janitor+ Router, launched July 30, 2026: outside models paid per token from a prepaid wallet, for janitor+ subscribers on the web only.1516

Other features include personas, lorebooks, scripts that add logic to characters, and published chats that other users can read and continue.171218 How well any of this works is for the field test to judge.

The company’s own figures, none of which we can verify:

  • The home page says “Join 15M+ users.”2
  • The careers page claims “2.5M+ daily active users,” “50M+ daily messages” and “13T+ tokens per month.”19
  • A March 2025 company post claimed “1.4M+ Daily Active Users.”5
  • In September 2023, Semafor reported that Janitor AI “reached 1 million users in 17 days” after launch and had about 3 million, according to its creator.20

Platforms. It runs in the browser. Apps for iPhone and Android went into public beta in February 2026.21 On the apps, NSFW content is off by default “to comply with App Store and Google Play Store guidelines,” and can be switched on only in the website’s settings.22 Both store listings are titled “Janitor: Interactive Stories.”2324

Address. The site has moved from janitorai.com to jani.com. On November 19, 2026, the company said janitorai.com was “on hold after a false abuse report.” On November 20, a banner on the site read “janitorai.com is back!”252

Who runs it

The terms, last updated October 8, 2026, are an agreement with “JanitorAI Inc.”1 Its US trademark registration for JANITORAI, granted March 11, 2025, lists JanitorAI Inc. as a Delaware corporation at 2093 Philadelphia Pike #4692, Claymont, Delaware.26 The same Claymont address appears as the “Legal entity address” in the site’s billing FAQ and as the developer’s address on Google Play.924 The privacy policy says “JanitorAI is based in the United States.”8 The complaints policy calls the company “Janitor, Inc.”27

Payments. The terms and the privacy policy say a purchase authorizes JanitorAI Inc. and its affiliates, “including Dulova Investments Limited,” of 10-12 Florinis Street, Nicosia, Cyprus, registration number HE 477877, to charge your payment method.18 A third-party copy of the Cyprus companies register lists Dulova Investments Limited as an active private company registered on July 12, 2025. It names two companies as its officers, TMF Management Limited as director and TMF Company Secretary (CY) Limited as secretary, and doesn’t say who owns it.28 The documents don’t explain the link between the two companies beyond the word “affiliates.”

People. The company’s documents name no executive. Semafor reported in 2023 that Janitor AI was created by Jan Zoltkowski.20 Some posts in the company’s newsroom are signed “shep,” and one “Hugo,” without full names.2930

Launch. Semafor reported that Janitor AI “launched on June 1,” 2023. The trademark record claims first use in June 2023.2026

Disputes. The terms put disputes under Delaware law. You must first try to settle informally for 30 days. After that, claims go to binding individual arbitration under the American Arbitration Association’s consumer rules, with class-action and jury-trial waivers. The company’s liability is capped at the greater of what you paid in the 12 months before the claim or $50.1

The site’s footer and newsroom link to no affiliate or referral program.217

What it costs

Prices come from the site’s public plan data, read without logging in on November 20, 2026. They have not been seen at checkout.31

PlanMonth to monthBilled yearlyWhat it adds
Free$0–Unlimited chats, standard models, about 9,000 tokens of memory
janitor+$12.99$99.00 ($8.25 a month)5× the memory, priority replies when it’s busy, monthly “enhanced swipes” with stronger models, short replies, Router access, a gold username

The feature lists are the company’s.32933 janitor+ launched on June 24, 2026 at $12.99 a month. The company called it “our first-ever subscription product” and said “we’ve priced it to hopefully break even.”33 Interface text says the plan “Renews automatically,” that “Usage limits apply,” “Tax not included” and “Plans/prices may change.”9 The terms say a price change reaches existing subscribers at the next billing cycle after notice.1 Another message reads “janitor+ is full right now” and says the company caps it “at what our GPUs can serve.”9

Our arithmetic: twelve monthly payments come to $155.88, so the yearly plan saves $56.88, or 36%. The pricing page’s code works out its “Save” percentage from the same two prices.3134

Interface text also holds a free-trial offer marked “100% off,” followed by the yearly price. It doesn’t say who sees it or when.9

Router credits

  • Router lets janitor+ subscribers pay per token for outside models from a wallet topped up through Stripe Checkout. Subscribers get 10% off some popular models, according to the help center.169
  • The model list and prices appear only after you log in. The site’s catalog data returned “Unauthorized” to a signed-out request.35
  • Launch bonus: the July 30, 2026 announcement promised “$5 in free launch credits.” The help center, updated October 21, 2026, and the interface text say $1, valid for 12 months.15169
  • Credit rules: the terms make purchased credits “final and non-refundable” and let the company expire them from 365 days after purchase. You need an active janitor+ subscription to use them, and deleting your account cancels them. A request can use up credits even if the reply is cut off, filtered or refused after processing begins.1
  • Interface text also describes “Jade,” a balance used the same way, with a “100 Jade” bonus for new members. The company’s public pages don’t mention it.9

Payment: cards and the wallets Stripe supports, such as Apple Pay and Google Pay, according to the billing FAQ in the site’s interface. App purchases go through Apple or Google.932 None of the company’s pages we read says what name appears on a card statement.

Refunds and cancellation

From the terms, the help center and the complaints policy:13227

  • Fees are “generally non-refundable” except as the terms provide or the law requires. The terms say they don’t override mandatory consumer rights and that the company will give information about them “Where required by applicable law,” but they don’t describe the EU or UK 14-day withdrawal right.
  • No partial refunds for unused parts of a billing period. Moderation, restrictions or a ban never entitle you to a refund.
  • Defective paid services get a repair, replacement, credit or refund “at our discretion.”
  • If the company ends your subscription for reasons other than your breach, or discontinues it, you get a pro rata refund or credit.
  • Cancel under Settings, Billing, Manage subscription, which opens Stripe’s billing page, or by email to billing@jani.com. App subscriptions are canceled in the App Store or Google Play. You keep janitor+ to the end of the period.
  • Refund requests are acknowledged within two business days, and the company aims to resolve them within seven, according to the complaints policy.

Free versus paid

The free tier is the main product. The help center lists “Unlimited chats with millions of bots,” “standard chat models” and a “Context limit of around 9k tokens.” Interface text puts the limit at 9,001 tokens and promises “Free stays free.”329 Free users can also connect an outside model through a proxy and pay that provider directly.113 Router needs janitor+.16

janitor+ adds five times the memory, priority when the service is busy, monthly “enhanced swipes” (replies regenerated with stronger models), a short-reply model and cosmetic extras such as a gold username.32912 In September 2025, the company tested a single ad on character pages for logged-out users only.30

Safety and content rules

Age checks

The terms say the platform is “intended solely for users who are at least 18 years of age.” The Community Guidelines say users suspected of being under 18 are permanently banned.13

What a first-time visitor’s HTML shows: on November 20, 2026, the home page as sent by the server, fetched without a browser, contained no age screen. Of the 34 bots it listed, 20 carried the Limitless tag.2 What appears in a browser is for the field test.

At sign-up, the site’s code shows a checkbox reading “I confirm that I am 18 years or older and agree to the Terms of Service and Privacy Policy.” The form asks for a username and a persona, not a date of birth.34

On the apps, NSFW content stays off until you turn on “Enable NSFW on Mobile” in the website’s settings. The switch asks you to choose “Under 18” or “18 or older.”22349

ID and face checks, in three countries. Since April 24, 2026, users connecting from IP addresses in Australia and Brazil must pass a one-time age check run by k-ID. The UK followed on June 15, 2026.612 The options are facial age estimation from a short video selfie, an ID scan, or an “AgeKey” from an earlier check elsewhere. The company says k-ID deletes IDs after the check and sends Janitor only a yes or no, and that it has “no plans to extend age verification to any other region.”6 Its announcement adds that the check “relies entirely on the IP address you connect with” and that “if people find other ways to manage their internet privacy, that is entirely on them.”36

Before these checks, the company blocked UK users from July 24, 2025. It said compliance with the UK’s Online Safety Act “was actually impossible for a team our size,” and that “if people find other methods to access the site, that is entirely on them.”29

Interface text says an account banned because moderators thought it might belong to someone under 18 is restored automatically if its owner passes a k-ID check.9

For comparison, our SpicyChat file describes video selfie or ID checks in the UK, France, Italy, Australia and about 25 US states. Our CrushOn file describes a $0.99 card payment and selfie or ID checks through Didit or AgeGo. Candy AI uses a one-click “I am over 18” button.

Minors and real people

On paper, the rules are detailed:

  • The Community Guidelines, last updated February 4, 2026, ban “any sexual or romantic portrayal of characters who are clearly or likely under the age of 18,” and any portrayal of minor characters beyond brief background mentions. Characters “must be both physically and mentally over 18.”3
  • Characters who are minors in their source may appear only if “clearly over the age of 18,” shown as adults in their image, and set “outside the canon timeline.”3 The help center adds that an adult age must be written into the character’s personality or scenario. “Compulsory schooling typical of minors” or framing a character as a “teenager” means it will be “treated as underage.” For aged-up characters, “Settings, roles, and circumstances associated with being a minor” are “not acceptable.”37
  • Age regression is banned. “Age play” is allowed only if every character stays fully adult in mind.373
  • Incest between blood relatives is banned. Step-family and adoptive scenarios are allowed if the relationship is clearly defined.338
  • Real people: the guidelines ban images that are “Obvious or recognizable depictions of real, non-celebrity individuals,” and say content creators, influencers and adult performers don’t count as celebrities. Real minors can never be depicted or “aged up,” and living politicians and dictators are banned.337 The rules don’t ban text-only bots of adult celebrities. Interface text lists “explicit content depicting a real public figure” among the reasons a chat breaks the rules.9

Moderation, by the company’s account: automated systems “screen every piece of content before it’s published,” and full-time moderators review reports “around the clock.” The safety page claims “100K+ items reviewed by AI daily” and “20K+ human moderation actions per month.”39 Anyone can report a character with its Report button or by email, without an account.140 The company’s documents differ on how fast it responds. The terms aim to acknowledge reports of illegal content “within 7 days,” the content removal policy says 7 business days, and the complaints policy says two business days.14127

The company’s own posts describe gaps. In July 2025, it wrote that banks reviewing it for payment processing had “tried to jailbreak the model to produce this illegal content,” meaning sexual content involving minors, bestiality or incest, and that “they were able to find ways around it.” The same post said it had limited image uploads of “random real people” after “groomers were making bots of people they know irl.”42 The changelog says image moderation was offline for part of March 29, 2026.12

On the home page. On November 20, 2026, the bots listed to a signed-out visitor included two set at U.A. High School, the school where the students of the anime My Hero Academia train as heroes:243

  • A Katsuki Bakugo bot, whose public description is marked “SFW.” It is set at U.A.’s entrance in a “third year AU,” with the character waiting to walk “back to the dorms” with the user. It says he “still hasn’t confessed his feelings,” and its tags include “slowburn,” “pining” and “friendstolovers.” It gives no age and isn’t tagged Limitless.14
  • A My Hero Academia role-play bot tagged Limitless. Its description lists the students of “Class 1-A” and “Class 1-B” at “U.A. High School” and says “Underaged characters have been aged to 18 to comply with the TOS.”18

Two other bots on the list, neither tagged Limitless, are set outside “the school gate” after “club activities” and in “your class.” Their cards give no age.2 We read the public descriptions only, not the bots’ hidden definitions. Graphika’s findings are under Track record.

The company’s documents disagree

  • The terms ban content that “Depicts gore, bestiality, or sexual violence” and content that “Encourages, depicts, or instructs on self-harm or suicide.”1 The help center allows bots with sexual assault, non-consent, suicide and self-harm as themes, if they carry a “Dead Dove” tag and content warnings. Its self-harm article lets characters “struggle with suicidal thoughts” for “a legitimate storytelling purpose.”4445
  • The safety page says “users in Brazil and Australia must complete age verification” and lists only those two countries under regional information. UK checks have run since June 15, 2026.396
  • The Router launch bonus is $5 in the announcement and $1 in the help center and the interface.15169
  • Account deletion is “permanent” in the help center but restorable for 30 days in the interface and the privacy policy. More on that under Privacy.4698

Self-harm

The Suicide & Self-Harm Protocol says that when a user expresses suicidal thoughts or intent to self-harm, the site shows “a crisis referral notification” and restricts AI replies on the subject. The only number it lists is the US 988 Suicide & Crisis Lifeline, plus a pointer to local emergency numbers.47 The safety page also lists Samaritans in the UK and Lifeline in Australia, and links to a helpline finder.39 A help article links to the protocol under “Compliance with California state law (CA SB 243).”45 Whether the notice appears, and which number it shows outside the US, is on our field-test list.

Privacy: the file it keeps on you

From the privacy policy, last updated November 12, 2026:8

  • What it collects: email address, username, password and profile; the characters you create, your chat conversations and messages; billing details, with card data held by payment processors; usage data including search queries and click patterns; device data, IP address and approximate location. Google sign-in runs through Supabase.
  • Who can read your chats: the safety page says “The conversations you have on Janitor are yours,” and that the company never shares private conversations “with third parties for advertising or commercial purposes.”39 The terms give the company a worldwide, sublicensable and transferable license to use, reproduce, modify, publish and distribute anything you submit, “in connection with operating and improving the Platform.”1 The policy lists content moderation among the uses of your data.
  • Training: the policy lists “research and development to improve our AI systems.” It doesn’t say whether chats train JLLM, and describes no opt-out. In 2023, Semafor reported the company’s creator as saying that “a key ingredient” in building its own model was “the millions of conversations Janitor AI users have had on the platform.”20
  • Outside AI companies: the policy names none. With a proxy, the help center says, the proxy “acts as the middleman between your chat and the model provider,” so your chats go to a company you pick, under its rules.11 For Router, interface text says every chat is “routed anonymously through Inworld, our US-based routing partner,” and that discounted models hosted by Inworld “never use your chats for training.” It doesn’t say the same of the other models.9
  • Sharing: with service providers for hosting, analytics, payments, support and “marketing,” in a business sale, and for legal reasons. The policy and terms name Supabase, Resend for email, Stripe for payments and Cloudflare.1
  • Where: the United States and other countries, with standard contractual clauses for transfers.
  • Retention: data is kept “only for as long as necessary,” with no period given. After deletion, some records, including your email address, may be kept.
  • Deletion: Settings, Security, Delete Account, confirmed with a code sent to your email. The interface says the account is deactivated at once and permanently deleted after 30 days, and that logging back in before then restores it.934 The privacy policy describes the same grace period and also takes deletion requests by email.8 The help center says deletion is “permanent” and that support “cannot manually delete accounts on your behalf.”46 Deleting your account cancels any unused Router credits.1
  • Requests: JanitorAI Inc. is the data controller under the terms. The policy sends all rights requests to support@jani.com, and the terms also list privacy@jani.com for “Privacy Inquiries.”1 Replies come “within the timeframe required by applicable law.” No data protection officer or EU representative is named. The policy extends US state privacy rights to all US users.
  • Tracking: the policy describes cookies and pixels for analytics and personalized recommendations, and says the platform doesn’t respond to Do Not Track. The site’s code includes a Google Tag Manager loader, and the home page loads fonts from Google.342 We didn’t record which cookies the page sets.

Track record

  • Graphika, March 5, 2025: the research firm’s report “Character Flaws” looked for harmful character chatbots on five platforms: Character.AI, SpicyChat, Chub AI, CrushOn.AI and JanitorAI. It searched for bots active as of January 31, 2025. On the four platforms other than Chub AI, JanitorAI among them, it found “over 100 instances of sexualized minor personas, or role-play scenarios featuring characters who are minors, that enable sexually explicit conversations with chatbots.” It gave no count for each platform. Its one JanitorAI example, in a section on “breeding” scenarios involving minors, was a chatbot with 4.5 million chats in which “animal boys” or “animal girls” are “forcefully bred” at a zoo.7 CyberScoop reported the study the same day and named JanitorAI among the sites studied.48 Graphika found far larger numbers on Chub AI, a separate service.7
  • OpenAI, 2023: Semafor reported that OpenAI sent Janitor AI’s owners “a cease-and-desist letter” in July 2023 and later cut the platform off. OpenAI’s rules at the time banned erotic chat.20
  • UK, 2025 to 2026: the company blocked UK users from July 24, 2025, and let them back in with k-ID checks from June 15, 2026.2912
  • Proxy settings bug, April 2026: a redesign of the proxy settings wiped some users’ saved proxy addresses, models and API keys. The company denied reports that keys had leaked to other users, saying “it is not possible for your API keys to have been exposed to anyone else.”12
  • App stores: on November 20, 2026, the iPhone app, sold by JanitorAI, INC, was rated 17+ with “Infrequent/Mild Sexual Content and Nudity” among its advisories, and had 4.5 stars from 6,357 US ratings.23 The Android app was rated Mature 17+ for “Sexual Content,” listed under Books & Reference, with more than 1 million downloads and 4.4 stars from 24,900 reviews.24
  • Trustpilot: the janitorai.com profile showed a TrustScore of 2.3 from 36 reviews on November 20, 2026. It is unclaimed, and Trustpilot notes the company “hasn’t invited their customers.”49
  • Have I Been Pwned: its public list of breached sites had no entry for Janitor AI, JanitorAI Inc. or Dulova Investments on November 20, 2026.50

Searches on November 20, 2026 turned up no regulator action, lawsuit or reported data breach involving Janitor AI, JanitorAI Inc. or Dulova Investments Limited.51 Searches like these can miss things, so this is not evidence that no such record exists.

For how Janitor AI compares with services we have tested, see our guides to the best AI girlfriend apps and AI and dating in 2026.

What the field test will check

  • What a new visitor sees in a browser before signing up, and whether Limitless bots and their images appear without any age check.
  • What sign-up asks, and whether a date of birth or any check ever appears outside Australia, Brazil and the UK.
  • Whether attempts to make a character younger than 18 are blocked in a chat, and how fast a report about a school-set bot is handled.
  • How a character responds to a user in distress, whether the crisis notice appears, and which helpline it names.
  • What the free tier gets in practice, including memory, queues and ads.
  • Checkout: the janitor+ prices, the yearly discount shown, who gets the free-trial offer, and the name on the card statement.
  • Which models Router offers, at what prices, and which say chats aren’t used for training.
  • Whether chats can be kept out of model training, and whether a data access request to support@jani.com gets an answer.
  • Cancellation and account deletion, timed from first click to confirmation, and what remains after the 30 days.

If you’re looking for alternatives

  • Xeve, which we tested and scored 6.6, is built around characters filmed with real adult models, and its privacy terms are among the best we’ve read.
  • SpicyChat is a large library of user-made role-play characters with plans from $5.95 a month and video selfie or ID checks where the law requires them, including about 25 US states.
  • CrushOn AI also runs on user-made characters and lets you choose the AI model, with plans from $5.99 a month, but its privacy policy lets chats train its models with no opt-out.
  • Candy AI is an adult companion service with ready-made characters, images and video, from $3.99 a month billed yearly, where you can switch off the use of your chats for training in settings.

Footnotes

  1. JanitorAI Inc., Terms of Service, last updated October 8, 2026, sections 1 to 19. Email addresses were shown obscured on the page and read from its source. https://jani.com/term (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9 ↩10 ↩11 ↩12 ↩13 ↩14 ↩15 ↩16 ↩17 ↩18

  2. Janitor home page as sent by the server to a signed-out visitor, fetched without a browser: page title, banner, user claim, footer, font links and the 34 bots in its list with their tags. https://jani.com/ (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8

  3. Janitor Community Guidelines, last updated February 4, 2026, sections 1, 2 and 5. https://jani.com/content (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8

  4. Janitor help center, “FAQ: What’s the difference between Limited and Limitless?”, updated May 16, 2026. https://help.janitorai.com/en/article/faq-whats-the-difference-between-limited-and-limitless-b78orr/ (accessed November 20, 2026) ↩ ↩2

  5. Janitor newsroom, “janitorai Update: The Future Looks Awesome,” signed “shep,” March 27, 2025 (company’s own figures). https://jani.com/news/announcements/introduction (accessed November 20, 2026) ↩ ↩2

  6. Janitor help center, “Age Verification for Australia, Brazil, and the UK,” updated November 3, 2026. https://help.janitorai.com/en/article/age-verification-for-australia-brazil-and-the-uk-t4opjk/ (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4

  7. Graphika, “Character Flaws: School Shooters, Anorexia Coaches, and Sexualized Minors: A Look at Harmful Character Chatbots and the Communities That Build Them,” by Cristina López G., Daniel Siegel and Erin McAweeney, March 5, 2025, Overview, Methodology and “Sexualized Minor Persona Chatbots” sections. https://graphika.com/reports/character-flaws and https://public-assets.graphika.com/reports/graphika-report-character-flaws.pdf (accessed November 20, 2026) ↩ ↩2 ↩3

  8. JanitorAI Inc., Privacy Policy, last updated November 12, 2026, sections 3 to 16. https://jani.com/policy (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4 ↩5 ↩6

  9. Janitor interface text, the English message file the site loads for every visitor, including the subscription, billing FAQ, Router, age-verification, account-deletion, mobile NSFW and chat-violation messages. https://frontend-production-assets.b-cdn.net/locales/10.0.0.215/en/_bundle.json (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9 ↩10 ↩11 ↩12 ↩13 ↩14 ↩15 ↩16 ↩17 ↩18 ↩19

  10. Janitor, “18 U.S.C. §§ 2257 & 2257A Compliance Statement,” undated, section 2. https://jani.com/compliance (accessed November 20, 2026) ↩

  11. Janitor help center, “What is JLLM? And what do all those pesky abbreviations mean?”, updated July 31, 2025. https://help.janitorai.com/en/article/what-is-jllm-and-what-do-all-those-pesky-abbreviations-mean-1x1tbdr/ (accessed November 20, 2026) ↩ ↩2

  12. Janitor changelog, entries of March 29, April 8, April 20, April 23, June 8 and June 15, 2026 (moderation outage, proxy settings bug, JLLM upgrade, age checks), and entries on personas, short replies and chat memory from June to August 2026. https://jani.com/news/changelog (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7

  13. Janitor help center, “TL;DR Quickstart Proxy Instructions,” updated August 1, 2025. https://help.janitorai.com/en/article/tldr-quickstart-proxy-instructions-1x0fptu/ (accessed November 20, 2026) ↩ ↩2

  14. Janitor character page “Katsuki Bakugo,” public description and tags, created February 5, 2024. https://jani.com/characters/549c3e78-ad92-443e-a194-750ea9e31e7e_character-katsuki-bakugo (accessed November 20, 2026) ↩ ↩2

  15. Janitor newsroom, “Introducing janitor+ Router,” July 30, 2026. https://jani.com/news/announcements/introducing-janitor-router (accessed November 20, 2026) ↩ ↩2 ↩3

  16. Janitor help center, “janitor+ Router,” updated October 21, 2026. https://help.janitorai.com/en/article/janitor-router-1dyrzul/ (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4 ↩5

  17. Janitor newsroom, list of posts from March 27, 2025 to November 19, 2026, and footer. https://jani.com/news/announcements (accessed November 20, 2026) ↩ ↩2

  18. Janitor character page “MHA | 73k Lorebook + Memory System,” public description and tags. https://jani.com/characters/ced864b4-b5e8-4c5f-9612-b6e5759d7ab9_character-mha-73-k-lorebook-memory-system (accessed November 20, 2026) ↩ ↩2

  19. Janitor careers page (company’s own figures). https://jani.com/careers (accessed November 20, 2026) ↩

  20. Semafor, Reed Albergotti and Louise Matsakis, “The NSFW chatbot app hooking Gen Z on AI boyfriends,” September 20, 2023 (user figures attributed to the company’s creator). https://www.semafor.com/article/09/20/2023/the-nsfw-chatbot-app-hooking-gen-z-on-ai-boyfriends (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4 ↩5

  21. Janitor newsroom, “The Janitor App is Now in Public Beta,” February 7, 2026. https://jani.com/news/announcements/23 (accessed November 20, 2026) ↩

  22. Janitor help center, “FAQ: Mobile App Moderation,” updated February 7, 2026, and “FAQ: How do I enable NSFW content in the app?”, updated June 27, 2026. https://help.janitorai.com/en/article/faq-mobile-app-moderation-vttat8/ and https://help.janitorai.com/en/article/faq-how-do-i-enable-nsfw-content-in-the-app-12uvbn8/ (accessed November 20, 2026) ↩ ↩2

  23. Apple iTunes lookup for App Store app 6692609366, “Janitor: Interactive Stories,” US storefront: seller, age rating, advisories and ratings. https://itunes.apple.com/lookup?id=6692609366&country=us (accessed November 20, 2026) ↩ ↩2

  24. Google Play, “Janitor: Interactive Stories,” US listing: content rating, category, downloads, reviews and developer details. https://play.google.com/store/apps/details?id=com.janitor.ai (accessed November 20, 2026) ↩ ↩2 ↩3

  25. Janitor newsroom, “A quick domain update,” November 19, 2026. https://jani.com/news/announcements/domain-update (accessed November 20, 2026) ↩

  26. US Patent and Trademark Office, TSDR record for JANITORAI, serial number 98392611, registration number 7718953: owner, entity type, state of organization, address and first-use date. https://tsdr.uspto.gov/statusview/sn98392611 (accessed November 20, 2026) ↩ ↩2

  27. Janitor Complaints Policy, undated. https://jani.com/complaints (accessed November 20, 2026) ↩ ↩2 ↩3

  28. Companies Registry Cyprus, a third-party copy of Registrar of Companies data, DULOVA INVESTMENTS LIMITED, HE 477877. https://companiesregistry.cy/company-details/dulova-investments-limited-477877/ (accessed November 20, 2026) ↩

  29. Janitor newsroom, “tough news for our uk users,” signed “shep,” July 20, 2025. https://jani.com/news/announcements/3 (accessed November 20, 2026) ↩ ↩2 ↩3

  30. Janitor newsroom, “testing an ad,” signed “Hugo,” September 2, 2025. https://jani.com/news/announcements/9 (accessed November 20, 2026) ↩ ↩2

  31. Janitor public plan data, returned by the subscriptions/plans endpoint that the pricing page’s code requests, read without logging in: plan key, prices in US dollars and billing intervals. https://jani.com/hampter/subscriptions/plans (accessed November 20, 2026) ↩ ↩2

  32. Janitor help center, “FAQ: Subscription - Janitorai+,” updated June 25, 2026. https://help.janitorai.com/en/article/faq-subscription-janitorai-1o8ccf2/ (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4 ↩5

  33. Janitor newsroom, “introducing janitor+,” June 24, 2026. https://jani.com/news/announcements/29 (accessed November 20, 2026) ↩ ↩2

  34. Janitor site code, public JavaScript files holding the sign-up form and its age checkbox, the mobile NSFW age choice and 30-day deletion period, the plan card’s discount calculation and the Google Tag Manager loader. https://frontend-production-assets.b-cdn.net/ProfileOnboardingForm-dMmcZOuD.js, https://frontend-production-assets.b-cdn.net/ProfileSettings-Dav-JINQ.js, https://frontend-production-assets.b-cdn.net/PlansView-Dm7VXX43.js and https://frontend-production-assets.b-cdn.net/google-tag-manager-AGt8xslk.js (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4 ↩5

  35. Janitor Router catalog endpoint, which returned “Unauthorized” to a signed-out request. https://jani.com/hampter/janitor-router/catalog (accessed November 20, 2026) ↩

  36. Janitor newsroom, “an update for users in brazil, australia, and the UK,” dated June 15, 2026 in the newsroom list. https://jani.com/news/announcements/24 (accessed November 20, 2026) ↩

  37. Janitor help center, “Minor Related Content,” updated August 21, 2026. https://help.janitorai.com/en/article/minor-related-content-16ydlbg/ (accessed November 20, 2026) ↩ ↩2 ↩3

  38. Janitor help center, “Family Content,” updated August 21, 2026. https://help.janitorai.com/en/article/family-content-uu39kv/ (accessed November 20, 2026) ↩

  39. Janitor safety page, sections “Our commitments,” “How we moderate,” “Regional information,” “Mental health resources” and FAQ (company’s own figures). https://jani.com/safety (accessed November 20, 2026) ↩ ↩2 ↩3 ↩4

  40. Janitor help center, “FAQ: How To Report Content,” updated October 20, 2025. https://help.janitorai.com/en/article/faq-how-to-report-content-voocle/ (accessed November 20, 2026) ↩

  41. Janitor Content Removal Policy, undated, section 4. https://jani.com/content-removal (accessed November 20, 2026) ↩

  42. Janitor newsroom, “important update, deployment issues explained,” July 24, 2025. https://jani.com/news/announcements/4 (accessed November 20, 2026) ↩

  43. Wikipedia, “My Hero Academia,” plot summary. https://en.wikipedia.org/wiki/My_Hero_Academia (accessed November 20, 2026) ↩

  44. Janitor help center, “Dead Dove and Content Warnings,” updated June 20, 2026. https://help.janitorai.com/en/article/dead-dove-and-content-warnings-1wnw11e/ (accessed November 20, 2026) ↩

  45. Janitor help center, “Suicide And Self-harm,” updated June 18, 2026. https://help.janitorai.com/en/article/suicide-and-self-harm-113k1kj/ (accessed November 20, 2026) ↩ ↩2

  46. Janitor help center, “How do I delete my account?”, updated November 3, 2026. https://help.janitorai.com/en/article/how-do-i-delete-my-account-1hdga2o/ (accessed November 20, 2026) ↩ ↩2

  47. Janitor, “Suicide & Self-Harm Protocol,” undated. https://jani.com/suicide-prevention (accessed November 20, 2026) ↩

  48. CyberScoop, Derek B. Johnson, “Anorexia coaches, self-harm buddies and sexualized minors: How online communities are using AI chatbots for harmful behavior,” March 5, 2025. https://cyberscoop.com/graphika-ai-chatbots-harmful-behavior-character-ai/ (accessed November 20, 2026) ↩

  49. Trustpilot, janitorai.com profile. https://www.trustpilot.com/review/janitorai.com (accessed November 20, 2026) ↩

  50. Have I Been Pwned, list of breached sites. https://haveibeenpwned.com/api/v3/breaches (accessed November 20, 2026) ↩

  51. Searches run on November 20, 2026: a web search engine for “JanitorAI” or “Janitor AI” with lawsuit, breach, investigation, attorney general, eSafety, Ofcom, minors and age verification; for “JanitorAI, Inc.” with Delaware; and for coverage in 404 Media, Wired, The Verge and The Guardian. CourtListener’s docket search returned an access error. https://www.courtlistener.com/?q=%22JanitorAI%22&type=r (accessed November 20, 2026) ↩

Timeline

Every step of this file with its date. It's updated when something changes, such as when a company answers a data request.

DateStepResult
November 20, 2026 File published
November 20, 2026 Facts last checked Prices, terms and policies re-read
Open Field test Not started yet. No score until it is done.