Janitor AI Safety Rating Index

Janitor AI scores a D (33/100) on the CompanionWise Safety Index, placing it in the Red tier. The score reflects a platform that grew rapidly on user-generated AI roleplay content but built almost none of the safety infrastructure that scale demands. For a quick overview, see the full Janitor AI review.

The most striking finding is a misleading privacy claim on the official safety page: “We never read your chats, and no chat content is shared with third parties.” Multiple evidence sources contradict this. The platform’s founder acknowledged using “millions of conversations hand-selected from users” to train JanitorLLM. Administrators can access chat logs for moderation and policy enforcement. And the bring-your-own-API-key architecture routes every message through third-party LLM provider servers. The claim may be narrowly true in a legal sense, but it misrepresents how user data actually moves through the system.

Age verification is where the platform fails most critically. Janitor AI’s own policy requires users to be 18+, but Google Play rates the app USK 12+, letting any 12-year-old download it. The only age check is a self-reported birthdate entry. A late-2025 attempt at mandatory ID verification triggered a mass user exodus and was effectively abandoned. Multiple child safety organizations flag Janitor AI as unsafe for minors, and the platform has zero parental controls, monitoring tools, or minor-specific safe defaults.

Crisis response is entirely absent. Despite facilitating emotionally intimate AI interactions at scale (approximately 2 million daily active users at peak), the platform has no suicide detection systems, no crisis hotline integrations, and no mental health safeguards beyond written content policies prohibiting depictions of self-harm.

Regulatory compliance is another floor-level score. Rather than comply with the UK Online Safety Act, Janitor AI blocked all UK users in July 2025, with the founder describing compliance as “impossible for a team our size.” User conversations were used for model training with no documented evidence of explicit consent, raising GDPR concerns.

On the positive side, Janitor AI’s tracker profile is cleaner than most competitors. The Exodus Privacy analysis found only two crash-reporting trackers (CrashLytics and Sentry) with no advertising or analytics trackers. Monetization is genuinely ethical: the platform is entirely free with no subscriptions, no paywalled safety features, and no manipulative upsell tactics. And the platform does not make therapeutic or mental health claims.

The core tension is a platform that collects minimal personal data and charges nothing, yet routes user conversations through opaque third-party infrastructure, lacks basic crisis safeguards, and cannot enforce its own age requirements. Safety infrastructure costs money, and Janitor AI’s free model leaves no budget to build it.