Is Replika Safe? Four Different Questions Behind One Word

“Is Replika safe” is four questions wearing one word, and they have different answers, different evidence, and different authorities. Is my conversation data handled in a way I would accept? Is the subscription arrangement one I can get out of? Is the app’s presence on my phone going to expose something to someone standing next to me? And is relying on it a good idea for me? Nobody can answer all four for you, and a page that says “yes, it is safe” has not distinguished them.

This post is about adults evaluating this app for themselves. Nothing here concerns anyone under eighteen, and that whole subject is outside this site’s scope.

Question one: data handling

The narrowest question and the one where the answer is most clearly located somewhere specific.

The operator’s published policy is the only authority on the operator’s practices. Not a review, not a comparison table, not this page. If you want to know what is retained, whether people ever read conversations, whether material is used to train models, and what deletion removes, those are statements a company makes in a document it maintains, and reading that document takes ten minutes.

Four sections are worth finding. Retention, meaning how long anything is kept. Human review, meaning whether employees or contractors can read conversations and under what conditions. Training use, meaning whether your text contributes to future models. Deletion, meaning what deleting an account actually removes and what survives it. If any of the four is absent from the document, the absence is itself the finding.

Policies have dates, and old dates matter. A policy last revised long ago on a product that has changed a great deal is a weaker assurance than a recent one, and the date is usually at the top.

Nothing about this is observable from outside. You cannot test retention. You cannot test whether a human read something. This asymmetry is permanent, and the practical consequence is that what you choose to send is a stronger control than anything you can verify.

Question two: the money

This one is fully knowable in advance and it is where avoidable harm actually happens.

Establish who you are paying before you pay. A subscription bought through a platform store leaves a third party holding the receipt, the cancellation screen, and a refund process that works whether or not the app cooperates. A card entered on an operator’s own site gives up all three. The distinction is set out in who you actually bought the subscription from.

Find the cancellation path before subscribing, not after. This takes a minute, it is the same minute either way, and it is much less pleasant to spend later.

Deleting the app does not end a subscription, which is the most common expensive misunderstanding in this whole category. Removing an app from a phone is a change to the phone; the billing arrangement was made somewhere else and continues undisturbed.

Question three: your device and the people near it

For a lot of adults this is the version of “safe” that actually bites, and it has nothing to do with the operator.

Notifications are the main exposure route by a wide margin. Whatever the app sends appears on a lock screen to whoever is nearby, and previews can be restricted per app. See notification previews and your lock screen.

The app’s existence is visible in more places than you would expect — the app list, device search, the store’s purchase history, an address-bar suggestion if you use the web version. On a device shared between adults, these are the practical concerns, and they are worked through in what a shared device exposes.

Screenshots and backups persist independently of the app. Both outlive uninstalling it.

None of this is specific to one product. It applies to any app you would rather not explain, and it is entirely under your control, which makes it the most tractable of the four questions.

Question four: whether relying on it is wise for you

The question people usually mean, and the one this page is least able to answer.

It is a question about your circumstances rather than about software, and it does not have a general answer that transfers between people. The site treats it separately and at more length in what to make of the mental-health claims, where it belongs. What is worth saying here is only this: it is a distinct question from the other three, and answering the technical ones tells you nothing about it.

Why there is no verdict here

Because a verdict would be worth less than the four distinctions above, and less honest.

The product’s behaviour is not fixed. Models, instructions and filters are server-side settings, changed without notice and without an app update. Anything specific written here about how the app behaves would describe one week.

This site has not tested it. No usage, no measurements, no inspection of traffic. A safety verdict on that basis would be a rewritten press release, and the category is already full of those.

The parts that matter most are not testable by anyone outside the company. Retention and review practice are commitments, not observable properties. The strongest honest position is to tell you where the commitments are published and what to look for in them.

So: the checkable questions are questions two and three, and you can settle both today. Question one is answerable by reading a document that is one link away from the store listing. Question four is yours. What is not available, from here or anywhere, is a single word that covers all four — and treating “is it safe” as one question is the actual mistake worth avoiding.