Replika's current terms gate romantic content behind a paid tier

The design

Two distinct Replika documents, read together, describe what the app is built to generate and what it withholds by default. The company's Terms of Service, retrieved 16 September 2026, states in its prohibited-content section that a user may not upload content that is 'unlawful, harmful, threatening, abusive, harassing, tortious, excessively violent, defamatory, vulgar, obscene, pornographic' or otherwise objectionable, and separately states in Section 1.1 that Replika 'makes no claims, representations or guarantees that the Services provide a therapeutic benefit.' A second document, the help-center article 'What is Replika Pro?', states that setting a companion's relationship status to 'Romantic Partner' and unlocking 'unblurred messages and photos, including romantic and creative selfies' are both features of the paid Pro subscription rather than the free tier.

What the evidence says

Neither page names the Italian Garante or the 2023 controversy directly; both simply state the app's current rules as they read today. A third document, Replika's own April 2023 engineering post, adds a layer the Terms do not: it says the company runs a classification system that tags every message, including ones the AI itself generates, into one of five categories, among them romantic and self-harm, and applies different filtering at each end of the exchange. Taken together, the three documents show a stated architecture of gating, through subscription and relationship-status controls on romantic content, sitting on top of a stated content-moderation layer, rather than one unified content-policy page.

What it asks of people

The design asks a user who wants a romantic-coded companion relationship to first select that status and, for image and message features, to pay for Replika Pro, which functions as a payment gate on the more intimate content the app can produce. It asks the same user to accept, without further detail, that a background classifier is scoring their messages and the model's replies against categories the company defines but does not fully publish the criteria for.

Privacy and safeguards

The Terms state that Replika is not a medical or mental-health service, that the app is restricted to users 18 and older, and that emergency situations should be taken to outside services rather than the app itself. None of the three documents states how long a message's classification label is retained, whether classification results are used beyond the immediate reply, or what independent review, if any, checks the classifier's behavior on romantic or self-harm content specifically.

• Does the current Terms of Service anywhere define what romantic content the classifier is trained to recognize?
• How does subscription status interact with the safety classifier described in the 2023 post?
• What would a reader need to see to confirm these three documents describe one coherent policy rather than three separate ones?

The available record is a stack of separately maintained documents, not one published content policy, and each states only what it states: legal prohibitions, a subscription feature list, and a description of a classifier the 2023 post says was already in use.