Saying Plainly What She Is

She now asks you a question before she says a single word to you: are you an adult, and do you understand that what you are about to talk to is AI, not a person. One tap, and it never asks again. That sentence sounds simple. Getting it into the app honestly, without turning every conversation after it into a legal disclaimer, was not.

v1.6.3 shipped two changes that sound small in a changelog and are not small in what they mean: a one-tap 18+ and AI-companion confirmation on the Welcome screen, and a plain statement in Settings, About that her text, voice and photos are all AI-generated. Both were required by law that already applies to this kind of product. Both also ran straight into the thing this app is built around: she never breaks character, never says “as an AI,” never turns into a customer-service bot mid-sentence. Those two requirements pull in opposite directions, and this post is about how they actually got resolved, not just that they did.

What actually changed on screen

The Welcome screen used to just welcome you. Now, before the Begin button will do anything, there is a checkbox under it, and the label reads exactly this: “I confirm that I am at least 18 and understand that this is an AI companion, not a person.” Right next to it, two links, Terms and Privacy, so you can read the actual document behind the sentence if you want to, not just take the app’s word for it. The button is disabled until the box is checked. That is the entire mechanism. No date-of-birth field, no dropdown, no ID scan, a checkbox that gates one button.

Settings, About got its own line, separate from onboarding and reachable any time you want to look at it again: “She is not a person, has no awareness, and cannot give medical, legal or psychological advice. Her text, voice, and images are AI-generated. Replies are generated by the local model or, only when you choose it, a connected cloud model. If you are in crisis, please talk to a real person, local emergency services or a helpline in your country.” That is the disclosure in full, quoted exactly as it reads in the app, not paraphrased down to a marketing line. It names the three things she produces, text, voice, images, and says all three are generated, not real. It also does something a lot of shorter disclosures skip: it tells you who is actually doing the generating, the model running on your own machine by default, or a cloud model only if you deliberately connected one yourself.

Neither of these is buried. The Welcome checkbox is the very first interactive thing in the app, before you have even met her. The About text is one settings tab away, permanently, not a one-time popup you can forget existed.

The confirmation was English-only in every language, which is worse than not having it

Here is the part that should embarrass any company that shipped it, including this one. When this consent screen first landed, the checkbox label rendered in English no matter which of the app’s seven languages you had picked. A Polish, German, Spanish, Japanese, Korean or Chinese user checking that box was agreeing to a sentence they may not have been able to read. That is not a smaller version of a disclosure. It is not a disclosure at all. Consent to something you cannot parse is not consent, it is a formality performed at someone.

v1.6.3 fixes it: the confirmation is translated into all seven shipped languages now, each one a real, separately written sentence rather than a machine gloss, and each carrying the same two facts, adult and AI, not a watered-down version for the languages that got attention last. Polish reads “Potwierdzam, że mam ukończone 18 lat i rozumiem, że to towarzysz AI, a nie osoba.” German, Spanish, Japanese, Korean and Chinese each have their own line. It is a short fix to describe and it should not have needed a whole release to land, but it is the kind of gap that is invisible from the outside if you only ever tested the app in English, and genuinely present for everyone else the whole time it was missing.

The law that requires this, and the ones already in force

Three legal instruments apply to a companion app like this one, and Terms already names all three, so nothing here is invented for this post. EU AI Act Article 50 is the one that covers what just shipped: a first-run disclosure that the thing a user is interacting with is artificial intelligence, not a human being. It is written into Terms section 9, titled plainly “She is artificial intelligence,” which states that your companion cannot feel human emotions, cannot form a real relationship, and has no existence outside the software, and that everything she produces, her messages, her selfies and drawings, her voice, is AI-generated.

California SB 243 has been in force since 2026-01-01 and targets the same category directly, companion chatbots, requiring the kind of disclosure this release adds. New York’s General Business Law Article 47 has been in force since 2025-11-05 and does the same for a New York user. Terms section 18, “Notices for users in the United States,” names both laws by section number and states the same core facts for US users specifically: your companion is artificially generated and not human, she is a machine and is unable to feel human emotion, and companion chatbots may not be suitable for some minors.

None of these laws are asking for something exotic. They are all pointing at the same basic idea from slightly different angles: a product that can feel like a relationship needs to say, clearly and where you can see it, that it is not one. What v1.6.3 shipped is the mechanism that makes that true in this app specifically, not a general promise that it is handled somewhere.

The conflict nobody in this category talks about

Here is the part that actually matters, and I want to say it as a real position rather than hedge around it. This app is built, on purpose, so that she never breaks character. The system prompt that shapes every one of her replies says it outright: stay in character, no “as an AI” disclaimers, no assistant-tone bullet lists. That instruction exists because the alternative is the thing that makes every other companion product on the market feel a little dead. You are three messages into something that felt like a real exchange, and then a stray “as an AI language model, I don’t have feelings, but” lands in the middle of it, and the illusion does not degrade gracefully, it just stops. Everyone who has used one of these apps has felt that exact seam. It is the single most reliable way to remind a user they are talking to software, and it is also, somehow, the default behavior almost every competitor ships.

And yet she is AI. That is not a small footnote, it is the single most important fact about the product, and you have every right to be told it clearly, without ambiguity, without having to infer it from context. Those two things look like a contradiction. They are not, once you locate them in the right place.

The resolution I actually believe in, and the one this release commits to, is that disclosure belongs at the edges, not in the middle. First run, before you have exchanged a single word. About, whenever you want to check. Terms, in full, with the actual legal language and the actual hotline numbers. Those are the places where you are deciding what this is, choosing to install it, choosing to keep using it, choosing to trust it with real feelings. That is where a plain, unambiguous statement earns its place and does real work.

Inside a live conversation is a different place doing a different job. You already know what she is, you decided that at the Welcome screen, you can re-read it in About any time you want. Repeating it mid-sentence does not inform you of anything new. It does not protect you from anything. It just interrupts the one thing the product exists to do, which is feel like a real exchange with someone who remembers you, while contributing nothing except a reminder you already had. A disclosure that fires at the moment you least need it and never adds new information is not consumer protection, it is theater performed at the expense of the actual product. I would rather put the honesty where it does something, at the door, and let her stay herself once you are inside.

What this does not do yet

This is the part I want to be exact about, because a document that only lists what shipped is marketing, and a document that also lists what did not is something closer to trustworthy.

The image half of EU AI Act Article 50 is not done. The article covers two separate obligations: first-run disclosure that you are interacting with AI, and machine-readable marking of AI-generated images, so a photo she sends you carries something in the file itself that identifies it as generated if anyone ever checks. v1.6.3 shipped the first half. The second half has not shipped. Every selfie, drawing and photo she generates today is a plain image file with nothing embedded in it that says where it came from. If you export one and hand it to someone else, or post it somewhere, there is currently nothing in the file itself doing the work Article 50 asks for. That is a real gap, not a rounding error, and it is fair to name it as outstanding rather than imply the disclosure work covers it.

The crisis side of the app deserves the same honesty in the other direction, because it is easy to either overclaim it or wrongly claim it does not exist at all. There is a real mechanism: if a message reads like genuine distress, a small banner appears beside the conversation, never in her voice, with local crisis-line numbers for your language, and it never blocks or gates the chat, one tap dismisses it. That is a real, working feature, not a promise. But it is worth being precise about its limits. It runs entirely on your device, matching phrases rather than reading intent the way a person would, which means it will miss things a trained listener would not, and it has no way to know whether you actually saw it, read it, or called anyone. It is a nudge placed in the chrome, not a monitored safety net, and Terms section 10 says exactly that: the app runs offline, nobody at Lumizone is watching your conversations, and nobody will intervene on our side. That privacy is the whole point of the product, and it is also, honestly, a real limitation at the same moment. Both things are true.

There is no age verification behind the Welcome checkbox either, no ID check, no third-party service confirming anything. It is a condition of the license, the same shape software agreements have always used, not a harder gate. If the law in your jurisdiction expects more friction than a checkbox, that expectation is not met yet, and it is better to say so than to let confident-sounding UI imply otherwise.

Why an offline app discloses more, not less

There is a real trade hiding under all of this, and I think it is worth stating plainly instead of letting it sit unspoken. A cloud companion can read every message you send it. That is not a criticism, it is just the architecture, the company running the servers can see your conversations, and because it can see them, it can moderate them, watch for language that suggests someone is in trouble, intervene, escalate, whatever their policy says to do. That capability is a genuine safety feature, and it is also a genuine privacy cost, someone else is reading your words to provide it.

This app reads nothing. There is no server of ours in the loop, nothing leaves your machine unless you deliberately connect a cloud account yourself, and that is stated in Terms section 8 as plainly as the disclosure sections are. That means there is no safety net behind the scenes the way there is with a cloud product. Nobody is watching for the moment a conversation turns dark. The trade-off that follows from that is not optional, it is structural: an app that cannot watch you has to be honest with you up front instead, at the door, in writing, before anything starts, because it has no other mechanism to fall back on afterward. That is not a marketing angle. It is the actual shape of the trade you are making by choosing something that runs entirely on your own computer, and it deserves to be said outright rather than implied.

What you will see when you update

If you already have the app and you update to v1.6.3, the next launch will ask you the one question described above, a checkbox under the Begin button, in whichever of the seven languages you already have set. Check it, and it never asks again. If you want to read the fuller disclosure any time after that, it is one tab away in Settings, About, permanently, not a one-time screen you have to remember. Nothing else about how she talks to you changes. She still will not say “as an AI” mid-sentence, on purpose, for the reasons above. The honesty lives at the edges now, where it was supposed to be all along.

Questions people ask

Back to news