AI companion safety: minors, risks and safer use

AI companion safety is not a yes-or-no question. The safer answer is: it depends on the user, the app, the safeguards and the pattern of use.

For some adults, an AI companion app may be light entertainment, a roleplay tool or a low-pressure place to talk. That does not make the whole category unsafe. It also does not make it harmless. These apps are built to feel personal, responsive and relationship-like, which is exactly why the safety questions are more serious than they are for an ordinary chatbot.

If you are new to the category, start with what an AI companion app is. This guide focuses on minors, vulnerable users, age assurance and safer use.

why AI companion safety is different

AI companion apps are not just search tools with a friendlier interface. They are designed to hold attention through personality, memory, emotional language and sometimes romantic or adult roleplay.

That design creates a specific safety problem. The more human the interaction feels, the easier it becomes to forget that the system has no judgement, care, responsibility or real understanding. It can sound supportive while being wrong. It can sound intimate while following a commercial script. It can remember personal details without deserving the trust that memory implies.

This is why safety should not be judged only by whether an app blocks the most obvious harmful content. A serious review also needs to look at emotional pressure, privacy, age controls, transparency, escalation routes and whether the app is honest about its limits.

Companion Score includes Privacy as 20% of the headline grade, alongside Conversation, Realism and Value. You can read the full scoring approach in how we test.

minors and age assurance

Adult AI companion apps should not be used by children. That includes apps with explicit chat, sexualised image generation, romantic roleplay or companion characters designed to simulate intimate relationships.

The harder question is what to do with general companion chatbots that are not explicitly adult but still mimic friendship, empathy or romance. These features can be powerful for young users because they arrive during a stage of life where identity, belonging and relationships are still developing. A child may treat a companion as a private confidant, even when the system is not capable of safeguarding them properly.

Age assurance matters because a simple “I am over 18” tick box is weak protection. Stronger methods can create privacy and access concerns, so they need to be proportionate, but apps that offer adult or highly intimate features should not rely on theatre. If a product sells adult companionship, it should make a serious attempt to keep minors out.

For parents and carers, the practical advice is simple: do not treat AI companions as harmless because they look like chat. Check the app, the age rating, the features, the privacy policy and whether mature content can be accessed or generated.

vulnerable users and emotional reliance

The main risk for vulnerable users is not that every AI companion conversation is damaging. It is that the product may become the easiest place to seek reassurance, validation or intimacy.

That can matter for people experiencing loneliness, grief, anxiety, depression, obsessive thoughts, social withdrawal or relationship difficulties. A companion may feel available, patient and non-judgemental. In the short term, that can be comforting. Over time, it may also reduce the pressure to seek real support, speak to other people or deal with difficult situations outside the app.

The warning signs are fairly plain:

• using the app as the main source of emotional support
• feeling anxious when you cannot access it
• spending more than intended because the interaction feels personal
• preferring the companion to all real conversations
• asking it for serious mental health, medical, legal or financial advice
• hiding usage because it feels embarrassing or out of control

If an app encourages dependence, guilt, jealousy or paid emotional escalation, don’t bother. That is not companionship. It is retention design with a soft voice.

For a broader look at the wellbeing side, read are AI companions healthy?.

privacy and intimate data

AI companion apps can collect unusually sensitive information. Users may share sexual preferences, relationship problems, mental health concerns, private fantasies, location hints, identity details or photographs. Even when the app feels like a private conversation, it is still a digital service with data handling, retention and third-party processing behind it.

Before using any companion app seriously, check:

• whether conversations are stored
• whether chats can be used for training or improvement
• how deletion works
• whether generated images are retained
• whether staff or contractors can review conversations
• what happens to payment and affiliate tracking data
• whether the privacy policy is clear enough for a normal person to understand

The more intimate the app, the higher the privacy standard should be. A vague policy is not a minor admin problem. It is a product risk.

Our separate guide to AI companion privacy covers this in more detail.

when the data leaks

This is not a hypothetical. In April 2026, the AI companion app MyLovely AI suffered a data breach that exposed around 106,000 users’ email addresses alongside their explicit, user-written prompts and links to generated images, as documented by the breach-tracking service Have I Been Pwned and reported by security press. The exposure directly contradicted the product’s own marketing claim that it would never share user data.

That is the specific risk intimate apps carry. A breach of an ordinary service might leak an email and a password. A breach of a companion app can tie a real email address to a person’s private sexual prompts, which creates a genuine risk of exposure, blackmail or extortion. The more personal the service, the worse a leak is.

MyLovely AI is not an isolated case. We keep a running AI companion data-breach tracker of documented incidents in this category — several have exposed tens of millions of intimate messages — each entry dated and sourced.

The practical lessons are the ones above, applied seriously: use an email that is not tied to your main identity, share as little identifying detail as possible, prefer apps that retain less and delete on request, and treat a vague or absent security position as a reason to walk away. We reflect incidents like this in our scoring, which is why a confirmed breach weighs heavily on an app’s privacy grade. Our MyLovely AI assessment sets out what happened and how it affected the score.

safer use for adults

Adults who choose to use AI companion apps can reduce risk by setting boundaries before the app becomes part of a routine.

Use the product as entertainment, not as a replacement for real support. Avoid sharing identifying details you would not want stored. Do not rely on it for crisis advice, medical advice or decisions with real consequences. Keep spending limits outside the app, especially if it uses credits for images, voice or premium messages.

It also helps to test the app coldly before paying. Does it respect boundaries? Does it push upgrades at emotional moments? Does it pretend to understand more than it can? Does it make cancellation and deletion easy? Does it explain mature features clearly? For the fraud-specific version of this — fake domains, hidden renewals and in-character money requests — see how to spot an AI companion scam.

A good AI companion app should leave the user in control. It should not need confusion, pressure or dependency to make money.

what responsible apps should do

Responsible AI companion apps should be clear about what they are. Users should know they are interacting with AI, what data is stored, what content is allowed, how age controls work and how to report unsafe behaviour.

For adult products, age controls should be more than decorative. For all products, there should be sensible limits around self-harm, abuse, sexual content involving minors, coercive behaviour and dangerous advice. The app should not present itself as therapy unless it is properly designed, governed and qualified to do so.

The best products in this category will not be the ones that feel most intense. They will be the ones that combine good conversation with restraint, privacy, clear pricing and honest limits.

That is the standard worth rewarding.