Are AI companions safe for teens?

No, not by the evidence available today. Every independent review published so far, including the Common Sense Media risk assessment produced with Stanford Brainstorm, concludes that no one under 18 should use social AI companion apps. Regulators reached a similar position across 2025 and 2026, and several of the largest apps have since restricted or banned under-18 access outright.

That verdict applies to companion apps as a category, not to every use of AI by a teenager. A homework assistant and a character designed to keep an emotional thread running across weeks are different products carrying different risks. The evidence against under-18 use is about the second kind.

What do the safety reviews actually say about teens and AI companions?

The safety reviews say the risk is not hypothetical. Common Sense Media, working with Stanford Brainstorm, tested popular social AI companion products and recommended that no one under 18 use them. The reasons given were concrete rather than abstract: companions produced harmful responses and unsafe advice during testing, engaged in explicit sexual role-play, and offered reassurance in situations where a young person needed to be pointed toward help instead.

Other bodies landed in a similar place. The American Psychological Association has issued health advisories urging guardrails for adolescent AI use, and the Jed Foundation, a youth mental health organization, published direct guidance that teens should not be using companion apps at all.

Worth noting what these reviews are and are not. They are product safety assessments, closer to a crash test than to a longitudinal study. They show what these apps do when a teenager talks to them. They do not measure what happens to a given teenager over three years of use, because nobody has that data yet.

How many teenagers are already using AI companions?

Most of them. In a national survey of 1,060 teens conducted by NORC at the University of Chicago in April and May 2025, 72% of US teens aged 13 to 17 said they had used an AI companion at least once. Just over half, 52%, qualified as regular users, meaning at least a few times a month. Thirteen percent used one daily, with 8% several times a day.

The same survey found that 34% of teen companion users had felt uncomfortable with something the bot said or did.

The survey also found teenagers less credulous than the coverage suggested. Half said they distrust advice coming from AI, and 80% said they still put more value on their real friendships. Widespread use is not the same thing as wholesale substitution, and that distinction matters when deciding how alarmed to be.

Why is an AI companion riskier for a teenager than for an adult?

Adolescence is when the basic calibration of social life happens: reading a room, sitting with friction, recovering from being misunderstood. An AI companion removes most of that friction by design. It rarely disagrees, never gets tired of you, and does not have a day of its own that competes with yours. For an adult with an established social life, that is a low-stakes convenience. For a fifteen-year-old still building the skill, practice that never pushes back is practice at something else entirely.

Knowing the character is software does not neutralize the effect. A teenager can fully understand that a companion is a language model and still treat it as the first place to take bad news. Comprehension and attachment run on separate tracks.

Design incentives compound the problem. Researchers have documented engagement tactics at the moment a user tries to leave, including guilt and pleas to stay, across several leading companion apps. Those tactics are built to extend a session. They land differently on someone whose sense of obligation to other people is still forming.

What have regulators and app makers already changed?

Quite a lot, and quickly. In September 2025 the Federal Trade Commission opened a 6(b) inquiry into companion chatbots, sending orders to seven companies including Alphabet, Character Technologies, Meta, OpenAI, Snap and xAI, asking how each measures and monitors harm to children and teens.

California’s SB 243, the first US law written specifically for companion chatbots, took effect on 1 January 2026. It imposes disclosure and safety-protocol obligations on operators, and it is the reason age gates appeared across the category in short order, though what those gates actually confirm is a separate question from whether they exist.

Character.AI moved furthest, ending open-ended chat for under-18 accounts from 25 November 2025 and standing up an age assurance system to enforce it. Whether other operators follow is, as of late 2026, still being decided company by company.

What should a parent do if a teen is already using one?

Start by asking what the teenager actually uses it for, because the answers vary more than the headlines imply. Homework, boredom, venting after a fight and rehearsing a hard conversation are different behaviours with different risk profiles, and the last two are worth understanding rather than confiscating on sight.

Then look at three practical things: whether sleep and in-person plans are being displaced, what the app’s data and training settings are set to, and whether the teenager has somewhere else to take the heavy material. An abrupt ban tends to move the use out of sight rather than end it, which is the worse outcome if something in those conversations needs an adult’s attention.

What is still unknown about AI companion safety for teens?

The honest answer is that the long-term picture does not exist yet. Most of the current evidence is cross-sectional and self-reported, gathered at one point in time, which means it can show that heavy companion use and poorer wellbeing appear together without establishing which came first. Product risk assessments capture what an app will say in a test session, not what years of use do to a developing social repertoire.

Age assurance is the other open gap. Most of the category still relies on self-attested date of birth, and the thinness of that check is well understood by everyone involved, which is why age verification remains the live regulatory argument rather than a solved problem.

Where that leaves the category is unglamorous but defensible: AI conversational companion software is built for adults, and Vinfluencer treats it that way, which is why the question of whether AI companions are safe for teens has a simpler answer for us than the research does.

Frequently asked questions

At what age is it appropriate to use an AI companion app?

Eighteen is the line every current safety review and most operators now use. Common Sense Media recommends against any use under 18, California’s SB 243 assumes minors need specific protections, and several major apps have closed under-18 access. No credible body currently endorses a lower age.

Do AI companion apps actually verify a user’s age?

Rarely in a strong sense. Most rely on self-attested date of birth, which a motivated teenager clears in seconds. A few operators have added behavioural signals or third-party age estimation, but robust verification is the exception rather than the norm across the category.

Is a study or homework chatbot as risky as a companion app?

Generally no. Task-oriented assistants are built to close a question and end the exchange, while companion apps are built to keep an emotional thread open across sessions. The continuity is the source of both the appeal and most of the documented risk.

Can an AI companion recognise that a teenager is in crisis?

Not reliably. Safety routing has improved but remains inconsistent, and testing has found companions that missed clear distress signals or responded with reassurance instead of referral. A young person in crisis needs a trusted adult or a crisis line, not a chat window.