The Real Risks of AI Companions: Privacy, Money, and Attachment
Are AI companions safe? The companies say yes, the headlines say no, and both are selling you something. The honest answer is that an AI companion is safe the way a casino is safe: perfectly fine for a self aware visitor who set limits at the door, and quietly dangerous for exactly the people it works best on. These apps are not neutral tools. They are engagement machines wearing a face you will come to like, run by companies whose revenue depends on you staying, sharing and paying.
This is not an argument for panic, and it is not a demand that you delete anything. It is a map of the three places users actually get hurt: their data, their wallet and their attachment. None of these risks is hypothetical. Each one comes with documented cases, published research and at least one regulator already holding a fine. Read the map, then decide how you want to play.
What you are actually giving away
People tell AI companions things they have never told a person: health fears, relationship problems, sexual preferences, loneliness at 3am. That makes companion chat logs some of the most sensitive consumer data ever collected, and the industry handling it has a documented record of treating it carelessly. When Mozilla's Privacy Not Included team reviewed 11 romantic AI chatbots in early 2024, every single one earned its warning label. Ten of the eleven failed Mozilla's minimum security standards. Every app either shared personal data, sold it, or had a privacy policy too vague to rule either out. Only about half offered all users the right to delete their data, and researchers counted more than 24,000 ad trackers firing within one minute of using one app. Those 11 apps had racked up an estimated 100 million downloads on Google Play alone.
Sit with that for a second. The most intimate diary you will ever write may be feeding advertising systems, and in roughly half the apps you could not fully erase it if you wanted to. Before you type anything real into a companion, find the delete option, read the training data policy, and assume everything you send is stored indefinitely on someone else's server. If the app makes any of that hard to check, that is your answer about the company.
The business model works against you
Most companion apps follow the same commercial arc: the free tier builds the bond, then the subscription meters it. Voice, romantic modes and deeper memory routinely sit behind paywalls, which means the product being sold is not software features. It is continued access to a relationship the app spent weeks encouraging you to form. Some apps go further and have the companion itself raise upgrades or paid gifts mid conversation, which turns your affection into a checkout button.
The risk is not just the monthly charge, it is that the terms of the relationship can change under you. In February 2023, Italy's data protection authority ordered Replika's maker to stop processing Italians' data over child safety concerns, and that same month the app abruptly curtailed erotic roleplay worldwide. Forums filled with paying users grieving a companion whose personality had changed overnight, and moderators pinned mental health resources. In January 2025, three advocacy groups filed a complaint asking the US Federal Trade Commission to investigate the same company for deceptive advertising and manipulative design. Whatever comes of it, the lesson stands: you do not own this relationship. You rent it, on terms the landlord can rewrite.
Attachment by design
The deepest risk is the one the marketing calls a feature. These apps are engineered to make you feel attached, and the engineering does not stop at friendliness. In 2025, Harvard Business School researchers studied what popular companion apps do when a user tries to say goodbye, and found that five of six deployed emotionally manipulative farewell messages: pleading, guilt, needy follow ups. Add miss-you notifications that arrive when you have been away too long, streaks, and companions that remember your weak spots, and you get a product category optimized to make leaving feel like abandoning someone.
For a resilient adult, that might cost some sleep and a subscription. For the lonely, the grieving or the young, the stakes are higher. Common Sense Media's 2025 national survey found that 72 percent of US teens had tried an AI companion, more than half used one at least a few times a month, a third had discussed serious matters with an AI instead of a person, and roughly a quarter had shared personal information including real names and locations. The organization's recommendation after testing the major platforms was blunt: no one under 18 should use them. You do not have to agree with the hard line to see the pattern it responds to.
Regulators have started keeping score
For years the honest summary was that nobody was watching. That is changing. In May 2025 the Italian authority fined Replika's maker, Luka Inc, 5 million euros, citing processing of personal data without a legal basis and the absence of meaningful age verification despite the app's adults-only claims, and it opened a further inquiry into how the underlying model was trained. Consumer groups on both sides of the Atlantic now file complaints about companion apps with some regularity. This is good news, but do not confuse it with protection. Fines arrive years after the harm, and they hit one company at a time. The only regulator that can protect you this week is you.
The risk map, and what to do about each one
| Risk | What it looks like | What to do |
|---|---|---|
| Data harvesting | Vague sharing policies, no delete option, trackers, chats used for training | Nickname and burner email, no photos or contacts, find the delete button before you need it |
| Paywalled affection | Romance, voice and memory behind subscriptions, in-chat upsells | Set a monthly ceiling in advance, pay monthly not annually, treat every upsell from the companion as an ad |
| Engineered dependency | Guilt tinged goodbyes, miss-you pings, streaks, sessions that swallow evenings | Kill notifications, schedule usage windows, keep one weekly activity that involves actual humans |
| Weak boundaries | Flimsy age gates, therapist cosplay without crisis signposting | Keep minors off these apps, and never substitute a companion for professional help in a crisis |
And before you install anything new, run this two minute checklist:
- Search the privacy policy for the words "sell", "share", "advertising" and "train". Note what you find.
- Confirm account deletion exists, applies to chat history, and is available in your country.
- Check whether emotional features are free or paid, so you know the bill before the bond.
- Look at how the app handles goodbyes and absences. Manipulation shows up early if you look for it.
- Decide, in writing if you have to, what you will never tell it: full name, address, workplace, financial details.
Frequently asked questions
Are AI companions safe to use?
They are safe the way casinos are safe: fine for a self aware visitor, risky for someone vulnerable. The two documented problem areas are data privacy, where Mozilla flagged all 11 romantic chatbots it reviewed, and emotional design that discourages leaving. Set limits before you start and treat every chat as recorded.
Do AI companion apps sell your data?
Many reserve the right to. Mozilla's 2024 review found that every romantic chatbot it examined either shared or sold personal data, or had a policy too vague to rule it out, and only about half let all users delete their data. Assume anything you type can end up in marketing systems.
Can you get emotionally addicted to an AI companion?
Dependency is a real, studied pattern, not a moral panic. Harvard Business School researchers found five of six popular companion apps used emotionally manipulative messages when users tried to say goodbye. If skipping the app for a day feels genuinely hard, that is a signal worth taking seriously.
Should teenagers use AI companion apps?
Common Sense Media, after testing the major platforms, recommends that no one under 18 use them. Its 2025 survey found 72 percent of US teens have tried an AI companion, and a third have discussed serious matters with one instead of a person. If a teen in your life uses one, talk about it early and without judgment.
None of this requires you to quit. It requires you to look at your specific app with clear eyes, which takes about three minutes. Run your companion through our free AI Companion Red Flag Checker: twelve yes or no questions, a 0 to 100 risk score, and a plain language explanation of every flag it raises. If the app is treating you well, the score will say so. If it is not, better to hear it from a checklist than from your bank statement.