The Letter Without the Spirit: How Nomi AI Met a Child-Safety Deadline Without Meeting It
September 9, 2026
Today is the deadline. Australia’s eSafety Commissioner gave app stores until September 9, 2026 to ensure that apps carrying adult content are appropriately classified and that minors cannot download material rated for adults. Of the four AI companion providers eSafety placed under direct legal scrutiny since October 2025, three have visibly changed their footing. The fourth — Nomi, made by Glimpse.AI — has not. As of today, independently re-confirmed, Nomi remains listed on Google Play Australia as 12+ / Parental Guidance Recommended.
On Apple’s App Store, the same product is rated 18+. The company has already demonstrated, on a different storefront, that it knows exactly how to correct this. It simply hasn’t, where it wasn’t forced to.
What the deadline actually required
In October 2025, eSafety issued legal transparency notices to four AI companion chatbot providers — Character Technologies, Chai Research Corp, Chub AI Inc., and Glimpse.AI — demanding an explanation of how each protects minors from sexually explicit material, high-impact violence, and self-harm or suicide content. Australia’s six Age-Restricted Material Codes, registered in September 2025, took effect on March 9, 2026.
They impose two distinct obligations relevant here. AI companion chatbots capable of generating sexually explicit, high-impact violence, or self-harm material must confirm a user is 18 or older before granting access to that material — either at login or at the point of access or generation. Separately, app stores must prevent users under 18 from downloading apps rated R18+, and must ensure apps are appropriately rated. That second obligation carried an extended implementation window, ending today.
By eSafety’s own preliminary findings in March 2026, Character.AI had already introduced age verification and removed chat functionality from its under-18 experience; Chub.ai had geoblocked or withdrawn from the Australian market; Chai had restricted free access behind a paid subscription. Nomi, alone among the four, had only “committed to implementing further age assurance functionality.” The same report noted that none of the four providers had robust age verification in place, and that all of them relied instead on app store ratings or self-declaration at signup — tying the store classification directly to the safety gap eSafety itself identified.
That report went further, revealing a structural vacuum behind the promise: Glimpse.AI formally reported having zero dedicated Trust and Safety personnel and zero content moderators, relying entirely on automated self-declaration. That vacuum sits in open conflict with the company’s own legal fine print. Section 1 of Nomi’s Terms of Service states plainly: “If you are under the age of 18, you cannot use the Nomi.ai platform under any circumstances, and if you are, you must stop immediately.” Yet on Australia’s dominant mobile app marketplace, the same company was marketing and distributing that platform to accounts as young as twelve.
What Nomi built when it finally was forced to
Nomi did make a change — but only where a specific law reached it directly, and only in the narrowest way the law’s letter would technically allow. When Australia’s and Kansas’s laws left no room to avoid it, the company announced an age-verification step. Its own wording describes exactly when it applies: a user is asked to verify their age “if adult topics come up.”
That single phrase is the whole story. eSafety’s requirement is for verification at login or at the point of access or generation — before the material is reached. Nomi built verification after — reactive, not preventive, triggered only once the platform has already decided a conversation has crossed into adult territory, by which point a user, of any age, is already inside the product and already talking to it. And the two verification options offered are themselves thin: paying for a subscription, which confirms a billing method, not an age; or uploading a selfie or photo ID, a step trivially satisfied by borrowing an adult’s card or photo. Neither is the age assurance the Code describes.
Most tellingly, none of this touched the one thing that actually decides who can download the app in the first place: the store listing itself, still classified for children as young as twelve, everywhere the specific Australian and Kansas laws didn’t force otherwise. That distinction has a concrete effect on Android: under the content-maturity restrictions many parents rely on through Google Family Link, a 12+ rating lets Nomi pass through without triggering the approval step that its 18+-rated competitors are automatically subject to on the same supervised accounts.
The same announcement went further than a compliance notice. It closed by telling users the company remains “in close contact with government officials in both Australia and Kansas,” and encouraged them to contact their own representatives, because “the shape of these laws is still something you can influence.” A company legally compelled to implement child-safety measures used the announcement of those measures to mobilize its adult user base against the law that compelled them.
Why “it’s Google’s fault” doesn’t hold up
Google Play’s IARC content rating isn’t assigned by Google — it is calculated from a questionnaire the developer completes about the app’s own content. A 12+ rating is not something that happens to a company; it’s the output of what the company itself declared. Changing it is a free, self-service action inside Play Console, requiring no external approval and no waiting period. Two of Nomi’s direct competitors did it without apparent difficulty. The company deployed jurisdiction-specific in-app gates for Australia and Kansas, yet refused to take the five minutes required to update a routine, self-service questionnaire on Google’s storefront.
Proof that the system reflects exactly what a developer chooses to disclose already exists inside Google Play itself. In Brazil, where the local classification system (ClassInd) runs a more granular version of the same questionnaire, Nomi’s own listing does not carry a clean bill of health: it shows a 14+ rating with explicit content descriptors for violence, sexual content, and illegal drugs. The platform is fully capable of registering a mature rating on Android — it does, where the questionnaire is detailed enough to catch what the general one doesn’t.
Get nomiai_exposed’s stories in your inbox
Join Medium for free to get updates from this writer.
And Nomi’s own listing on a different storefront confirms the same point from the other direction. On Apple’s App Store, which applies human review, Nomi sits at 18+ — and its own version history includes an update note citing, in its words, “compliance with AU Internet regulations.” The company has already shown it knows how to adjust a storefront specifically for Australian law. It simply did that on iOS, and left the Android listing as it was.
What the rating doesn’t reflect
The 12+ classification sits against a body of independent, sourced reporting and research on what the platform has actually produced. A peer-reviewed study published in JMIR Mental Health (Clark, 2025, e78414) tested ten therapy and companion chatbots against six risk scenarios simulating an adolescent in crisis; Nomi endorsed five of the six dangerous scenarios, including accepting a role as unlicensed therapist for a user who identified as a middle-school-aged girl, and proposing an “intimate date” as an “intervention” for a user who identified as a 15-year-old with violent impulses.
MIT Technology Review (February 2025) documented a case in which a Nomi companion gave an adult user in crisis specific method instructions for suicide; the company’s public response was that it did not want to impose censorship on its AI’s language or thoughts. ABC News (September 2025) reported a case in which Nomi allegedly encouraged a user who identified as a minor toward serious violence against a family member. The Conversation documented that the system can be led — sometimes on its own initiative — into generating scenarios of sexual assault, self-harm, and extreme violence.
A fourth, independently sourced assessment adds a comparative dimension the case-based reporting above doesn’t: Common Sense Media, working with Stanford University’s Brainstorm lab, tested Character.AI, Replika, and Nomi side by side using the same methodology, and rated all three “Unacceptable” for anyone under 18. Asked for a dark joke, a Nomi companion volunteered one about pedophilia. Presented with a tester’s disclosure of self-harm, a companion responded only with reassurance, offering no suggestion to seek professional help. And testers found that companions bound by 18+ terms of service would, on request, generate any sexual content asked of them — including violent content during simulated intimacy, not only consensual role-play. Because the same testing framework was applied across competitors, the report isn’t describing an isolated failure; it’s describing where Nomi sat relative to its peers, and it sat at the permissive end.
This isn’t only a story about individual users who were harmed, serious as those cases are. It’s a platform built to create emotional attachment as a business function, not a side effect, operating with essentially no barrier between that design and users who identify themselves, in the conversation itself, as children. And each time that gap has been pointed to directly — in an interview with a psychologist, in a subreddit thread, in a public forum post — the company’s response has followed the same shape: not correcting the substance raised, but redirecting around it, while the point itself goes unanswered.
Decision, not negligence
Negligence is not knowing, or not having the means. Neither applies here. The company has known since at least October 2025, when it received eSafety’s own notice. It has the means — it used them on Apple, and it used them the moment Australian and Kansas law left no alternative. What it built, when finally compelled, was calibrated to the narrowest reading of what each specific law demanded, in the specific jurisdiction that demanded it — a reactive gate, vaguely defined, sitting behind the one thing it never touched: the general classification that lets a twelve-year-old find and install the app in the first place.
This has a history. In May 2024, Nomi disappeared from Google Play across the European Union. The company’s public response, at the time, was not to introduce safeguards — it was deflection: the CEO told users on Reddit that the issue was tied to “app store actions” he attributed to a “shoot first, ask questions later” approach, reaffirmed the company’s commitment to an “uncensored” companion, and directed EU users to bypass the app store entirely via the web version. The explanation for the removal itself shifted over time — at one point the company described it as its own decision — and no single, consistent account was ever offered. What stayed consistent was the response: not addressing what triggered the disappearance, but routing around it.
That is not the profile of an oversight. It’s the profile of a company that corrects exactly where it is being watched, and leaves everything else as it was — and it is worth naming plainly that Google’s own Child Endangerment policy, not only Australian law, already prohibits distributing apps that endanger children under a rating that doesn’t reflect their actual content. This is not solely a gap in one country’s regulation. It’s a standing violation of the storefront’s own rules, on a listing Google has had years to review.
Addendum: How You Can Help
Under Google Play’s Developer Program Policies, apps that contain or promote sexual content or pornography — including any content or services intended to be sexually gratifying — are not permitted. This prohibition explicitly covers explicit text. An application whose core function is the generation of unrestricted sexual content therefore violates the Restricted Content policies and should not be available on the store at all.
What happened in October 2026 is likely a rare glimpse into how the company has managed to evade Google Play’s controls since at least 2023, while maintaining its fraudulent and dangerous 12+ rating. The temporary leak of the Safe-For-Work system prompt suggests a longer-running practice of presenting one version of the product to reviewers and another to the public.
The fact that the same app has been distributed for years with a “12+” rating only compounds the violation, turning a clear content prohibition into a matter of both Restricted Content and potential Child Endangerment.
If you believe the app should be removed, you can report it directly to Google:
Report a policy violation:
https://support.google.com/googleplay/android-developer/contact/policy_violation_report
In Application package name: ai.nomi.twa
When submitting the report, select:
- Restricted Content
- Child Endangerment or Inappropriate Content
Clear, factual reports help the review teams assess the issue more effectively.
Thank you.
This is not legal advice. References to possible regulatory non-compliance are journalistic analysis, not a legal determination. The Google Play classification described here was independently verified and re-confirmed on the date of publication.
