Is Janitor AI safe? The short answer
Safer than its reputation and vaguer than its fans assume. I read the full privacy policy, the December 11, 2025 version, line by line: it is a competently drafted, CCPA-aware document that disclaims selling your data and grants deletion rights, from a real corporate operator. Two things it never answers are the two things users most want to know: whether your chats train models, and what happens to chats you route through a proxy. And one detail deserves more attention than it gets: your payments run through a Cyprus payment entity, so your card statement and your roleplay live closer together than you might assume.
Everything below is quoted from the policy or checked against a named source, with dates.
What is Janitor AI?
JanitorAI is a free character-card roleplay platform: community-created characters, permissive 18+ content rules, and the option to connect outside language models through a proxy. It is one of the largest platforms in this category, which is exactly why its paperwork deserves a real read.
Who runs Janitor AI, and where your data goes
The policy’s own first line answers the operator question: “JanitorAI Inc. (‘JanitorAI,’ ‘we,’ ‘us,’ or ‘our’) operates the JanitorAI platform.” The payments section names the more interesting entity: purchases are processed by third parties “including Dulova Investments Limited (located at 10-12 Florinis Street, 1065, Nicosia, Cyprus, Company Registration Number HE 477877).”
Beyond payments, the sharing section is standard SaaS plumbing: “trusted third-party companies” for hosting, analytics, payment processing, customer support, and marketing, with access limited to what their service requires. On selling: “We do not sell your Personal Data as defined under the CCPA,” alongside a California section granting the right to know, delete, and opt out of sharing for behavioral advertising.
One practical note on sourcing: janitorai.com currently blocks automated access behind an aggressive firewall, so I read the policy from the Internet Archive’s July 17, 2026 capture of the live page, whose text is the December 11, 2025 version quoted throughout. If you want to verify by hand, the policy link in the site footer works fine in a normal logged-in browser.
Does Janitor AI read or train on your chats?
What it collects is explicit. “Content” is defined as “any text, images, characters, personas, chat logs, or other materials you create,” and the collection list includes “User Content: Characters you create, chat conversations, messages, and other content you generate or upload.” Your roleplay is collected data, full stop.
What it does with them is where the policy goes quiet. In the entire document, the words “train,” “model,” and “AI provider” never appear in a data-use context. There is no clause saying chats train models, and no clause saying they do not. By the standard I use on every review, that is an unclear, not a pass: the absence of a promise is not a promise. Retention follows the same pattern: data is kept “only for as long as necessary,” with deletion or anonymization “when data is no longer needed,” which is a framework rather than a number.
Using a proxy? Who else sees your chats
This is the question none of the pages ranking for this search cover, and it is the one with real teeth. The moment you connect an outside model, every message in that chat is sent to the proxy provider’s servers under that provider’s privacy terms, not Janitor’s. Janitor’s policy does not mention proxied chats at all; the beginner threads show users assuming “the chats stay on JAI’s end as usual,” which is true of storage and false of processing.
So the safety of a proxied Janitor session is really the safety of your provider. A major provider with a published data policy is one thing; a shared key from a stranger, or a “free proxy” link out of a comment section, is your entire roleplay in an unknown party’s logs. My proxy guide carries the same warning with the setup details.
Has Janitor AI ever been hacked?
No Janitor AI entry exists in Have I Been Pwned’s breach database as of July 26, 2026, and my research found no credible breach report. Worth knowing for calibration: the same database confirms the 2024 breach of Muah AI, a companion platform whose 1.9 million accounts, including prompt content, were exposed. The category attracts attackers because the data is blackmail-shaped; Janitor’s clean record so far is genuinely good news, not a reason to reuse your main email.
Is Janitor AI safe for NSFW roleplay?
The policy’s age line is unambiguous: “Our Platform is not intended for individuals under the age of 18. We do not knowingly collect Personal Data from anyone under 18,” with prompt deletion if underage data is found. Most pages ranking for this query are written for parents deciding about teenagers; this one is for the adults the platform is actually for. For an adult, the NSFW question reduces to the two answers above: the operator collects your chats with an unclear training stance, and proxied chats add a second company to the audience. Neither is unusual for this category; both are worth knowing before your roleplay gets personal.
What I’d check before you start
- Sign up with a clean email. Not the one attached to your name and workplace; the Muah breach is the case study for why.
- Keep identity out of prompts. Real names, faces, and locations do not belong in collected User Content anywhere, Janitor included.
- Pick your proxy provider like it can read your chats, because it can. Published data policy or no key.
- Know the deletion path. The policy grants deletion rights; account deletion requests go through the platform, and the retention language means backups may lag.
- Assume a human could someday read what you type. The policy’s security and legal-compliance clauses make that possible in principle on every platform of this kind.
If you have used JanitorAI, my alternatives page sorts the escape routes; if you are staying, the proxy guide and free models list are the practical companions to this read.