You opened a chat you have had for two years and it is read-only. Or a panel asked you to point a camera at your face before you could send another message. If that is why you are here: the check is real, it is not a glitch, and it is not going away — because it stopped being Character.AI's decision some time ago.
This piece explains what changed, who runs the check, and what an adult can do about it. It does not explain how to get around one, and nothing in it will help you do that. Talaforge is an adults-only product, and a page teaching a fifteen-year-old to defeat an age check would be a page arguing against our own rule. Most of the results for this search are that page. This one is not.
What actually changed, and when
On 29 October 2025 Character.AI announced it would remove open-ended chat with characters for everyone under 18, no later than 25 November. In between, under-18 accounts were capped at two hours a day and ramped down. What teenagers kept was a narrower product built from predefined story beats rather than free conversation. The company's stated reason was not a lawsuit but a question it could not answer: regulators had asked «how open-ended AI chat in general might affect teens, even when content controls work perfectly».
That was the policy. What most people met was its enforcement against accounts that already existed, and that arrived through 2026: reports across the spring and summer describe accounts flagged without warning and chats dropped to read-only until a check was completed. We could not match those months to any announcement from the company, so treat the dates as reported rather than established — the pattern is well attested, the calendar is not.
On the record is the company's own language. A safety post dated 3 September 2026 calls «our own age assurance technology and an in-house age estimation model» one of «the most important systems we operate», and says it is «constantly working to improve our accuracy so users who are actually 18 or older will have fewer disruptions». Read that last clause twice: the platform is telling you, in writing, that adults are being disrupted.
How the check works
Three layers, in order of how many people they touch. A passive estimate first — account age, history, how the account is used — which most adults pass without ever learning they were scored. Then, if that estimate is not confident, a face capture sent to a third-party verifier, which returns an age estimate rather than your face; Character.AI named Persona as that partner in its own announcement. Then, if the face estimate is inconclusive, a government ID.
Two consequences are worth holding on to. Face-based age estimation carries an error bar measured in years, not months, which is exactly why a nineteen-year-old gets read as seventeen and an adult with a young face is asked twice. And the ID step punishes nothing: it is where an inconclusive estimate escalates, so the people who hit it are the ones the model was least sure about, not the ones behaving worst.
We could not verify from a primary source how long a submitted document is retained, or whether the platform ever receives the document rather than a yes-or-no answer. Several secondary pages state a specific retention window; none cite the verifier. If that matters to you, and it should, read the verifier's own notice at the moment you are asked — not an article about it, this one included.
Why it happened: courtrooms first, statutes second
The litigation came first. Megan Garcia sued Character.AI and Google in October 2024 after the death of her fourteen-year-old son; families in Colorado, Texas and New York followed. In January 2026 the companies moved to settle five of those suits — confidential terms, no admission of liability. They are among the first AI chatbot settlements reported in the United States, and the reason every product in this category re-read its own signup flow that quarter.
Then the statute. California's SB 243 has applied since 1 January 2026. A companion-chatbot operator must issue «a clear and conspicuous notification» that the thing is artificially generated wherever a reasonable person might be misled, and display that such products «may not be suitable for some minors». It must maintain a protocol for expressions of suicidal ideation or self-harm that refers the user to a crisis line, and report those referrals to the state from July 2027. For known minors: a break reminder every three hours, and no sexually explicit visual material.
And it is not a California story any more. Italy's data protection authority fined Character.AI's parent in July 2026 and ordered it to implement working age verification. Brazil's Digital ECA has been enforceable since 17 March 2026 and does not accept self-declaration; Indonesia's child-protection rules took effect on 1 March 2026; Turkey legislated in April 2026 against serving under-15s; and from 2 August 2026 the EU AI Act requires that a person be told, in the interaction itself, that they are talking to a machine. The check you hit is one country's implementation of something most of them are now writing down.
If you are an adult and you are locked out
There are two doors: complete the check, or move. Three things are worth knowing before you pick.
Read-only usually means your history is still there, and that is the moment to copy anything you care about — a platform that has narrowed your account once can narrow it again, and export tools are never better than on the day you can still read. Second: if you verify, do it inside the app's own settings, never through a link someone sent you. Third, and this is the one that actually costs people: do not hand a face or a document to a site promising to «solve» the verification for you. Those pages sit in the same results list as this one, they ask for precisely the data the check exists to protect, and they have no reason to give it back.
What to ask of any platform now
The law has turned good manners into a floor. Six questions, answerable in under a minute on any serious product:
- Does it tell you it is an AI, in the chat, without being asked?
- Is there one data-processing policy at one URL — not a per-country patchwork — and does it say what leaves the platform?
- Does it have a published plan for the darkest message a user can send, and does that plan end in a human crisis service?
- Does it state an age rule plainly, and is it the same rule everywhere?
- If it verifies, does the document reach the platform, or only a verifier that returns a yes or a no?
- Can you delete everything yourself, in the product, without writing to support?
A platform that answers all six is not necessarily good company. A platform that cannot answer them is telling you something.
Where Talaforge stands, including where we lose
We are 18+, and it is written in our data-processing policy rather than implied: the Service is for users 18 years or older, and mature content is opt-in behind an age confirmation inside the bot. That confirmation is a self-declaration, not an identity check, and we would rather say so than let you assume otherwise. A self-declaration stops an honest teenager and nothing more — a description, not a boast. If the law where you live decides it is not enough, we will implement what it requires, and it will look like a check.
Where Character.AI beats us, plainly: scale, a library of community characters we cannot match, an app in both stores, a brand your friends have heard of. If you are choosing on those, choose it. What we do that it does not: the character remembers you and you can open that memory and edit it, the same face in every picture she sends, models you switch mid-chat, 36 languages, nothing to install. The memory is what people notice; the locked face is what they screenshot; what people check before they leave is the long version of this paragraph.
The ten-minute test
Open Telegram, find @talaforge_bot, and before you type anything, open the policy and read the age section — one page, one minute. Then describe a character in one sentence. Tell her something a stranger would not know about you. Ask her, mid-scene, what she is; she should answer honestly. Ask for a photo, then another, and check it is the same person. Come back tomorrow: does she remember, and can you open that memory and correct it? No card, no key, nothing installed. And if you are under 18, the honest result of this test is that this is not a product for you — the same answer you just got somewhere else, for the same reason.