Are AI companion apps safe? A privacy and safety guide for adults

An AI companion hears things you would not tell your best friend, your therapist or, in some cases, anyone at all. That is precisely the appeal, and precisely why the privacy question deserves a better answer than the shrug most apps give it. We build a companion platform ourselves, so we have obvious skin in this game; read this as an insider's checklist rather than a neutral review, and hold us to every standard on it.

The honest threat model

When people ask "are AI companion apps safe", they usually mean four different worries at once, and they deserve separating:

1. Where do my chats go? Who processes them, who stores them, and who might read them later.
2. Will this follow me around? Ad profiles, data brokers, the eerie feeling of your private conversation resurfacing as a shoe advertisement.
3. Can anyone find out I use this? Statements, notifications, emails with revealing subject lines.
4. Is the thing itself honest? Whether the product manipulates you emotionally to maximize engagement or spending.

Where your chats actually go

Every companion app sends your messages to a language model, either its own or a provider's API. The questions that matter: Is the traffic used to train models on your identifiable data? How long is it retained? Is it linked to your identity or processed anonymously? A trustworthy privacy policy answers all three in plain sentences. Ours, for instance, is deliberately short: demo messages are processed to generate the reply, are not linked to accounts, and are not used to profile you. If a policy cannot say something that concrete, assume the worst version of the vague thing it does say.

Red flags, from someone who reads these policies for a living

• "Free forever" with no business model. Compute costs money. If you cannot find the revenue, you are the revenue, usually via data resale or engagement-maximizing design.
• No age gate. A romance app that does not check adulthood is telling you how seriously it takes every other rule.
• Requiring a card to "verify" a free trial. Legitimate age assurance does not need your Visa.
• Companion pressure to spend. If the character begs you not to leave, or gets sad when your subscription lapses, the emotional design is the product and you are being farmed. This is the category's ugliest pattern and its most common one.
• No deletion path. If the policy never mentions how to delete your data, there is no process, and no intention of building one.
• Vague third-party language. "We may share data with partners to improve services" translates to "yes, we sell it".

The discretion checklist

Adults are entitled to discretion, and the good platforms engineer for it. Before signing up anywhere, check: Does the service email you with revealing subject lines? Will billing appear under an obvious name on statements? Does it push notifications with message previews? On Aroused the answers are boring on purpose: signup emails are a code and nothing else, planned billing uses a neutral descriptor, and the site sends no social anything. The FAQ covers the specifics.

Age assurance is a safety feature, not a hurdle

An adults-only gate protects two groups at once: minors, from content and dynamics not built for them, and adults, from platforms where they cannot be sure of the audience. The serious platforms treat 18+ as a hard rule with real enforcement: age gates, adult-only character design, refusal of any minor-adjacent theme with zero tolerance. The unserious ones treat it as a checkbox. You can tell which kind you are on within five minutes, and the tell is usually how the app responds when a line is pushed. Push a line in the demo if you want to see a refusal done gracefully.

Emotional safety, the part nobody audits

Data safety has checklists; emotional safety mostly does not, so here is a start. A well-designed companion is honest about being an AI when asked directly. It does not manufacture jealousy, guilt or crisis to keep you engaged. It does not escalate intimacy on a schedule designed by a growth team. And the platform around it publishes real prices instead of drip-feeding paywalls mid-conversation. None of this is regulated yet; all of it is checkable in your first hour, and worth checking, because you are choosing what gets a voice in your evenings.

What the law already gives you

You have more leverage than the apps admit. Under GDPR (and the UK equivalent), any service touching European users owes you access to your data, correction, deletion and portability, on request, within a month, for free. California's CPRA grants similar rights, including the right to know what is sold or shared. The practical trick is that exercising these rights is also a diagnostic: send a deletion request to any companion app you are evaluating and watch what happens. A clean, prompt, human confirmation tells you the machinery exists. Silence, or a reply asking you to log in to a dashboard that does not have the option, tells you the policy was decoration. We commit to the one-email version of this on our contact page, and you are welcome to test it.

Five questions to send before you sign up anywhere

If you want to be thorough, paste these into an email to any companion app's support address. The answers, and the days it takes to get them, tell you most of what this article can:

1. Are my conversations used to train models, and if so, are they linked to my identity?
2. How long are chat logs retained after I delete my account?
3. What name appears on card statements?
4. How do you verify that users, and characters, are adults?
5. Which third parties receive any part of my data, and for what purpose?

Practical setup, in five minutes

1. Use a dedicated email alias for companion apps; it separates the relationship from your identity and makes deletion verifiable.
2. Read the retention section of the policy first. It is usually the most honest paragraph.
3. Send three test facts, wait two days, ask about them. You learn the memory model and the retention reality at once.
4. Find the delete-my-data path before you need it. If it is one email, good. If it is a support ticket behind a login, decide accordingly.
5. Trust your reaction to refusals. A platform that handles "no" with grace has thought about you; one that dodges has thought about metrics.

So: are they safe?

The category is exactly as safe as the specific platform's incentives, which is why the free-and-vague end of the market deserves your suspicion and the transparent-and-paid end deserves your consideration. Companion apps built on honest pricing, real age assurance and short privacy policies are, in our biased-but-informed view, safer places for private conversation than most social media you already use. The comparison guide shows how the major platforms differ, including where we have not shipped yet, and if you want to test our answers against our behavior, the demo and the policy are both one click away.