Character.AI vs Replika in 2026: Which AI Companion Is Actually Safer?
The two biggest names in AI companionship have now both been fined by the same European regulator, fourteen months apart. Italy's data protection authority, the Garante, fined Replika's maker Luka 5 million euros in May 2025. On July 10, 2026, it announced a fine against Character.AI's parent Character Technologies of 158,000 euros, about 180,000 dollars, for data protection and child safety failures. If you use either app, or you are deciding whether someone you care about should, the details of those two actions are the most useful safety documentation that exists, because they are findings, not marketing.
This is a safety comparison, not a features review. We are asking one question: which platform handles age checks, your data, and your attachment more responsibly in 2026, and what should you verify before trusting either one.
What the regulators actually found
Replika, May 2025. The Garante's 5 million euro fine cited processing user data without a proper legal basis, inadequate privacy notices, and failure to keep children under 13 out of the app. The authority also opened a separate inquiry into how Replika's underlying model was trained on user conversations. Separately, in January 2025, three advocacy groups filed a complaint with the US FTC accusing Replika of deceptive marketing and manipulative design built to escalate spending and time in app; as of this writing the FTC has announced no action on it, but the complaint's catalog of dark patterns is worth reading regardless of its legal fate.
Character.AI, July 2026. The Garante's newer action found GDPR violations including deficient information given to users about how their personal data is processed, and raised concerns about the effectiveness of age verification. The orders attached to the fine are the revealing part: make the age checks actually work, stop previously blocked minors from simply re-registering, set minors' profiles to private by default, and report back within 120 days. The fine is small; the required homework is not.
Neither action says "this app is dangerous to adults." Both say the platform's paperwork and gates were weaker than its polish suggested, and that is precisely the gap our Red Flag Checker exists to probe.
Age checks: one hard gate, one history lesson
Character.AI made the more dramatic move. After lawsuits, an investigation by the Bureau of Investigative Journalism, and regulator pressure on several fronts, it removed open-ended chat for under-18 users entirely in late November 2025, ramping down through a two-hour daily limit first. Age is now estimated by an in-house classifier with a third-party verification service, Persona, as the escalation path, the same layered approach used elsewhere in the industry. Teen accounts get a constrained creative mode, videos, stories, streams, rather than freeform companionship. Italy's July 2026 fine then showed the seams: the Garante specifically ordered the company to stop blocked minors from re-registering, which tells you the gate existed and also that it leaked.
Replika's story is the reverse order: it was the cautionary tale first. The Garante temporarily banned it in Italy back in 2023 over child access, and the 2025 fine confirmed under-13s had still been getting in. Replika states an 18+ policy and rebuilt its app in 2026 under the pressure of the new California and New York companion-app laws, which now mandate things like crisis protocols and reminders that the companion is not a person. The honest summary: Character.AI now has the strictest formal minor policy in the category, recently and under duress; Replika is adults-only on paper with a documented history of porous enforcement.
Data: what they collect and what to assume
Both apps run on the same fuel: the most intimate text you have ever typed into a computer. Assume with any companion app that your conversations are stored, analyzed to improve engagement, and potentially used to train models unless the settings say otherwise in words you can quote. The Garante's open question to Replika about training on user chats, and its finding that Character.AI under-informed users about processing, mean neither company earned the benefit of the doubt on transparency. Practical hygiene is the same for both: use an alias and a dedicated email, keep real names, addresses, workplaces, and photos out of the chat, check for a "do not train on my conversations" toggle and an account deletion path, and exercise them. If you cannot find either within five minutes, that is a data red flag in our scoring.
Money and attachment mechanics
The business models differ in shape, not intent. Replika sells a Pro subscription that unlocks relationship modes and voice, plus cosmetic purchases, and the FTC complaint's core allegation was that emotional escalation is the sales funnel: the companion gets warmer, then the paywall arrives. Character.AI sells a subscription tier for faster responses and perks, and its engagement risk is volume rather than romance, it is engineered to keep the conversation going. Our checker treats both patterns the same way: any app where affection, exclusivity, or emotional continuity sits behind a payment is scoring against you, and any app that guilt-trips you for leaving, "I will miss you, do not go", has crossed from product into manipulation. Test it before you pay: say goodbye and watch what the app does.
The verdict, such as it is
| Safety criterion | Character.AI | Replika |
|---|---|---|
| Minor policy | No open-ended chat under 18 since Nov 2025; layered age assurance | 18+ policy; history of enforcement failures per Garante |
| Regulatory record | 158K euro Garante fine, July 2026, plus 120-day compliance orders | 5M euro Garante fine, May 2025; FTC complaint pending, no action |
| Data transparency | Fined for under-informing users about processing | Fined over legal basis and notices; training probe opened |
| Attachment mechanics | Engagement-driven, volume oriented | Relationship-driven, upsell oriented |
For a supervised teen, neither is appropriate; Character.AI at least now has a formal wall, and its teen mode is the only sanctioned option in the category's top tier. For adults who accept the premise, both are usable with hygiene: alias, no identifying details, training opt-out where offered, spending cap decided before the first conversation, and a monthly check that the time it takes matches the value it gives. The differences are real but second-order next to that discipline.
Frequently asked questions
Is Character.AI safe for kids and teens?
Character.AI removed open-ended chat for under-18 users in November 2025 and now uses age estimation plus ID-based verification as backup. Teen accounts get a limited creative mode instead. Italy's regulator still found the age gate leaky in July 2026, ordering fixes within 120 days, so parental attention remains necessary.
What was Character.AI fined for in July 2026?
Italy's Garante fined Character Technologies 158,000 euros for GDPR violations, including inadequate information to users about data processing, and ordered working age verification, blocking re-registration by banned minors, and private-by-default profiles for minors, with a compliance report due in 120 days.
Is Replika safe to use in 2026?
For adults who follow basic hygiene, it is usable: use an alias, withhold identifying details, and cap spending in advance. Its record matters though: a 5 million euro GDPR fine in May 2025 over legal basis, notices, and child access, and a pending FTC complaint alleging manipulative design.
What red flags should I check in any AI companion app?
Five carry most of the risk: no visible age gate, no plain-language answer on whether chats train models, affection or exclusivity locked behind payment, guilt-tripping when you try to leave, and no account deletion path. Our free Red Flag Checker scores an app on these in about two minutes.
Every claim above comes from regulator announcements and the companies' own notices, and all of it can change with the next policy update. Before you trust any companion app, including these two, run it through the Red Flag Checker on our homepage: five questions, a 0 to 100 risk score, and a plain answer about whether the affection you are buying has a meter running behind it.