If you have assumed these apps verify age the way a UK porn site now has to, they do not, and the reason is more interesting than simple negligence.
What most apps actually do
A date-of-birth field, or a button that says “I am 18 or over”. That is it. No document check, no facial age estimation, no card verification. Entering a date in 2001 takes about two seconds and is not checked against anything.
Of the sixteen apps in our review set, these are the ones whose age assurance we have checked most closely:
| App | Age assurance |
|---|---|
| Joi AI | Age assurance for UK users, added after Ofcom investigation |
| Character.AI | Age signals and substantial under-18 restrictions, introduced after legal scrutiny |
| Replika | Self-declared, tightened after Italian regulatory action |
| SweetDream AI | Self-declared 18+ |
| Kindroid | Self-declared, plus app-store rating |
| Nomi AI | Self-declared |
| Candy AI | Self-declared |
| EVA AI | Self-declared, plus app-store rating |
| FantasyGF | Self-declared |
| Muah AI | Self-declared |
One genuine check in ten. Two if you are generous about Character.AI, which is a different sort of case.
The rule is narrower than the headlines suggest
UK readers are understandably confused, because since July 2025 pornography sites serving the UK have had to use highly effective age assurance, and the change was very visible. Yet apps generating explicit images of a companion you designed ask only that you tick a box.
The usual explanation is that chatbots fall outside the Online Safety Act. That is only half right, and the half that is wrong matters.
Ofcom’s position is that a standalone chatbot can sit outside the Act’s core scope where all three of the following hold: it does not connect users to one another, it does not search multiple websites or databases, and it cannot generate pornography. Ofcom has also said that chatbots which can generate pornography already fall within the Act’s age-assurance duties.
That last condition is the one that bites. An app whose entire proposition is generating explicit images of a companion is not a service that “cannot generate pornography.”
Applying that test to the apps we reviewed
We cannot test the first two conditions from the outside. We can test the third, because it is a product feature: does the app generate images, and does it apply a content filter?
Fourteen of the sixteen apps we reviewed generate images with no content filter. On Ofcom’s stated test, those fourteen appear to be within scope for age-assurance duties. Thirteen of them ask only that you tick a box.
The two that fall outside on this test are Replika, which removed adult roleplay in 2023, and Character.AI, which does not generate images and filters content hard.
So the position is not “the rule does not reach them.” For most of this category the duty appears to exist and is simply not being met. That is a different problem with a different solution, and it is why enforcement rather than legislation is what has actually moved things so far.
We could not retrieve Ofcom’s pages directly — their site blocks automated access — so this is drawn from reporting of Ofcom’s published position rather than quoted from Ofcom. Read Ofcom’s online safety publications before relying on it, and tell us if it has moved. This is general information, not legal advice.
We are not the only ones finding this
The Ada Lovelace Institute published a market scan of the UK AI companion sector in January 2026. Of the 17 top UK platforms it analysed, 15 had weak or non-existent age verification as of July 2025, including sites hosting explicit content. It singled out Candy.ai for showing sexualised avatars on landing pages reachable by any child with a browser.
Its conclusion was blunt: “Our review shows a widespread failure of age verification on AI companion sites that are accessible in the UK.”
Our sample is smaller and our method is different, and we arrived at the same place.
What is coming
The government is preparing rules that would set a minimum age of 18 for romantic companion chatbots — those designed to simulate sexual relationships or roleplay — with similar restrictions for under-18s on intimate features more broadly. Reporting indicates the regulations are expected to be laid before Parliament before the end of 2026, and their final form is not settled.
Several sites state that Ofcom has confirmed mandatory AI chatbot age verification by 2027. We could not find that confirmed by Ofcom or by government, and we are not repeating it.
The exception, and what it tells you
Joi AI is the one app here that checks properly, and it did not arrive there voluntarily.
Ofcom opened an investigation into Novi Ltd, the provider of the service, concerning compliance with age-check requirements. Reporting indicates the investigation was closed after the company introduced age assurance for UK users. We have flagged that closure as reported rather than confirmed, since we could not verify it against Ofcom’s enforcement record.
The outcome is what matters. Joi AI is a mid-table product on our overall rubric — it scores 7.5 — but it is the only app here scoring above 8 on safety and age assurance. If age checking is your deciding factor, it outranks apps that beat it overall, which is exactly the sort of case our single-criterion rankings exist for.
The broader lesson: regulation in this space has so far moved one company at a time, through investigation, rather than through a rule applying to everyone.
Character.AI is a different kind of case
Character.AI has faced significant US litigation and regulatory attention over harm to minors, and has since introduced substantial restrictions on under-18 accounts, including limits on open-ended companion chat.
Those are real changes, more substantial than anything else in this category. But the history is a company being forced to act rather than acting first, and that is reflected in its score: it rates 9.0 on value and 7.0 on safety.
It is, nonetheless, the most likely of these apps for a teenager to encounter, because it is free, it is heavily discussed, and it does not present as an adult product.
If you are a parent
Honest assessment: the apps will not do this for you. Three things that actually work, roughly in order of effectiveness.
Device-level controls. Screen Time on iOS and Family Link on Android restrict by app-store age rating and block browser categories. More effective than anything app-side, because it operates at a layer a teenager cannot talk their way around.
The payment barrier. Almost everything meaningful in these apps sits behind a card. A household where card details are not saved to the device is a household where the free tiers are the realistic ceiling.
The conversation. More useful than either, and worse advice to give because it is harder. These apps are compelling for teenagers for understandable reasons — they are patient, available and never embarrassed. Approaching it as a security problem to be solved tends to work less well than approaching it as a thing worth talking about. Survey reporting in 2026 suggested around one in five UK teenage boys had encountered AI companion apps, so this is not an unusual conversation.
If you are an adult user
Two things worth knowing.
Weak age checks are a signal about everything else. An operator not investing in age assurance is usually not investing in security, moderation or data handling either. It correlates with the apps that also do not name an operating company.
It may change. Operators tend to build to the strictest regime they face and ship it everywhere. Australia’s industry codes explicitly name AI companion chatbots and include age assurance duties, and the EU AI Act’s transparency obligations began applying in August 2026. UK users often inherit rules written elsewhere.
Several sites state confidently that Ofcom has confirmed mandatory AI chatbot age verification by 2027. We could not find that confirmed by Ofcom or by government, and we are not repeating it.
The summary
- One app in ten genuinely checks, and only after a regulator made it.
- Most are not currently obliged to, because of where the Online Safety Act’s scope falls.
- If this matters in your household, Joi AI is the one that checks.
- Device controls and payment friction do more than any app setting available to you.