Are AI companion apps safe? Privacy, age rules, money and scams explained (2026)

Need to talk to someone now? If you or someone you know is in crisis, contact local emergency services or find a free helpline at findahelpline.com. In the US you can call or text 988; in the UK, Samaritans are on 116 123.

How this page was made. This is a documentation-based guide. It pulls together AITried’s app-by-app safety guides, each built from the company’s own privacy policy, terms and help centre, plus regulator and researcher publications. We have not audited any app’s security. Every policy quote below carries the date we read it.

The short answer

AI companion apps are reasonably safe for adults who use them with care, and not built for children. The risks are real but ordinary: your conversations are stored and sometimes used to train models or sent to outside AI providers; age checks are often just a tick-box; subscriptions renew automatically and refunds are rare; some people come to rely on a companion more than they meant to; fake clone sites take payments; and companies can fail or be sold, taking your data with them. None of these is a reason to panic. All of them are reasons to read one page before you sign up.

The six real risks, and what to do about each

1. Your chats are data

Every major companion app stores what you write. The differences are in what else happens to it:

• Character.AI keeps chat content and uses it to improve its models unless you opt out (full guide).
• Replika’s privacy policy (read 26 October 2026) says it “will never use or disclose the content of your Replika conversations for marketing or advertising purposes” (full guide).
• Janitor AI’s safety page (read 26 October 2026) says “We never sell your personal data or share private conversations with third parties for advertising or commercial purposes”, but if you plug in an outside model through a proxy, your messages go to that provider under its own rules (full guide).
• Candy AI’s notice says chat content may go to third-party AI providers and de-identified chats may be used for training, with possible human review (full guide).
• Nomi’s privacy policy (read 26 October 2026) says “We do not and will not sell or rent to any third party any of your personal information”, and that deleting your account deletes personal information “within 28 or so days”, except material in training archives, which would “no longer be attributable to you”.

Researchers have found real weaknesses in this category. In February 2024 the Mozilla Foundation reviewed 11 romantic AI chatbots and found 10 failed its minimum security standards, and most did not let users opt out of having chats used for training. In September 2025 the US Federal Trade Commission ordered seven chatbot companies to explain how they use conversation data and enforce age limits. Italy’s data protection authority fined Replika’s maker EUR 5 million in 2025. Use our 12-point privacy checklist before you share anything personal.

2. Age rules and how they are checked

Replika, Nomi, Kindroid, Janitor AI and Candy AI all set a minimum age of 18. How they check differs: Character.AI uses age assurance through a third party (a selfie first, ID only if needed, per its help centre); Janitor AI requires age verification through k-ID only in some countries (Australia, Brazil and the UK) and says users suspected to be under 18 “may be suspended or banned”; Candy AI relies mainly on a self-declared age gate. Character.AI is the only app here that accepts teen accounts, and since November 2025 under-18s no longer get open-ended chat. If a teen has been locked out, our guide to Character.AI’s age checks explains what happened. Parents can link to a teen’s Character.AI account to see screen time, but not messages.

3. Money

The most common harm users report is financial, not technical. Subscriptions renew automatically. Refund windows are short or absent: Candy AI’s terms allow refund requests within 24 hours of payment, card payments only; Nomi’s terms say “There are no refunds or credit for partial months of service”; App Store purchases follow Apple’s refund process. Several apps charge tokens or credits on top of the subscription for images, voice or video. Kindroid ended unlimited free chat in early October 2026 and 12 of its 50 newest US App Store reviews (2 to 24 October 2026) are complaints about it, which shows how quickly the deal can change. Our guides to cancelling an AI subscription and what AI companions really cost cover the details.

4. Emotional reliance

Companion apps are designed to be pleasant to come back to. Many people find them comforting; some find they are spending hours a day or pulling away from friends. App Store reviews show both: a recent Character.AI review describes using it “every single day for hours” for four years, while a Replika reviewer writes that it “has helped me so much”. If a companion leaves you feeling worse, lonelier or anxious when you are away from it, take a break and talk to someone you trust, or a helpline. Treat what an AI says about your health, money or relationships as a conversation, not advice.

5. Scams, clones and fake payment pages

Popular app names attract copycats: clone sites and apps that look official, take a payment and deliver little or nothing. The safe habits are simple: type the official address yourself, check the developer name in the app store listing, and never pay through a link in a message, comment or ad. Our guide on how to spot fake AI app clones lists the warning signs.

6. When the company is sold or shuts down

Character.AI filed for Chapter 11 on 12 October 2026 and is running a court-supervised sale. It says the platform keeps running and its privacy practices are unchanged, but it has not said whether chats are part of what is sold. Nomi’s privacy policy says plainly that if it sold the company, “your personal information would be transferred to the successor or purchaser”; most policies have a similar clause. The protection in your control is a data export: Character.AI’s help centre says one can take 3 to 4 weeks, so request it early. See what happens to your Character.AI chats in the sale and how to read the export file.

How the main apps compare on safety (checked 26 October 2026)

“See guide” means the answer needs more than a cell can hold; the linked guide quotes the policy.

A five-minute safety routine before you sign up

1. Find the official site or app by typing the address or checking the developer name, not through an ad.
2. Read three things in the privacy policy: whether chats train models, whether they go to outside AI providers, and how deletion works.
3. Check the age rule and the refund rule in the terms.
4. Pay monthly, using a payment method you can see clearly on a statement, and write down the renewal date.
5. Keep chats free of identifying details about yourself and especially about other people.

Verdict

AI companion apps are safe enough for adults who go in with open eyes: assume the company can read and keep what you write, pay monthly, use only official sites and apps, and know how to cancel. They are not a place for children, and they are not a replacement for people or professional help. Before you choose one, read its safety guide above; the differences between apps are bigger than their marketing suggests.