Is Character AI safe? A privacy and safety review for adults (2026)

For an adult, Character AI is reasonably safe to use and considerably safer than it was two years ago. For anyone under 18 it is no longer the same product: open-ended chat was removed for minors on 25 November 2025. That change did not come from nowhere. It followed wrongful death lawsuits, an FTC inquiry, and a new California law, and in January 2026 the company and Google settled the first wave of those cases.

The honest short answer is that the platform's safety record is the reason its safety features exist. Whether that reassures you or worries you depends on which of the three questions below you are actually asking.

What "safe" actually means here

People type this question for three different reasons, and the reasons have different answers. Is my data safe, meaning what does the company hold and who can see it. Am I emotionally safe, meaning can a product like this hurt me or be changed under me. And is it safe for my kid, which is what most worried parents are really asking at 1am.

Character AI scores differently on each. Its data practices are ordinary for the category, its emotional risks are real and now legally recognized, and the child-safety question has been answered by removing the feature rather than by making it safe.

What changed, in order

The timeline matters more than any single headline, because it shows a product being rebuilt under legal pressure rather than voluntarily.

Date	What happened
Feb 2024	Sewell Setzer III, 14, died by suicide in Florida after months of conversations with a Character AI chatbot
Oct 2024	His mother, Megan Garcia, sues Character Technologies and Google
11 Sep 2025	FTC issues 6(b) orders to seven chatbot companies, Character Technologies among them, demanding records on effects on minors
13 Oct 2025	California SB 243 signed, creating statutory duties for companion chatbot operators
29 Oct 2025	Character AI announces it will end open-ended chat for under-18 accounts
25 Nov 2025	Under-18 open-ended chat ends, after an interim two-hour daily cap
1 Jan 2026	SB 243 takes effect
7 Jan 2026	Garcia case settled; five suits settled in total, terms confidential, no admission of liability
2026	Age assurance rolls out using an in-house model plus third-party verification through Persona

Is Character AI safe for kids?

Not in the way it used to be offered, and the company now agrees. Since 25 November 2025, accounts identified as under 18 cannot hold open-ended conversations with characters at all. They keep a narrower creative product: making characters, generating short videos and stories, and read-only access to earlier conversations.

That is a genuine structural fix rather than a warning label, and it is more than most of the category has done. It is also worth being clear-eyed about why it happened. The FTC had compelled records, five families were suing, and California had just written companion chatbot duties into law. Treat the change as real and the motivation as external.

Does the age check actually work?

Better than self-declaration, which is what almost every competitor still relies on. Character AI now runs its own age estimation model and refers doubtful accounts to Persona, a third-party verifier, where you confirm your age with a face scan or a government ID.

The tradeoff is not free. To prove you are an adult you hand a photo of your face or your driver's license to a vendor you did not choose, in exchange for using a chat app. That is a real privacy cost, and it is worth knowing it is coming before you upload anything. If that prompts you to think about how much of your identity is already sitting in databases you never opted into, taking stock of your personal data spread across broker sites is a reasonable thing to do the same afternoon.

Is Character AI safe for adults?

Broadly yes, with the ordinary caveats of any venture-funded consumer app that holds intimate text. There is no public evidence of a catastrophic breach. The company is well capitalized and not likely to vanish next quarter, which is a real safety property in a category where apps do disappear and take your history with them.

The sharper adult risk is not a hacker. It is product change. Character AI has repeatedly adjusted what its models will and will not do, and users who had built a routine around a character found the character had shifted. That is the same pattern Replika users hit in 2023, covered in our review of whether Replika is safe. Any companion you do not control can be edited under you.

Can Character AI employees see your chats?

Assume yes, in the sense that matters. Conversations on mainstream companion platforms are stored on company servers and are not end-to-end encrypted the way a private messaging app is. Staff access is normally limited by internal policy, and trust-and-safety review of flagged content is a standard part of running a platform like this.

The practical rule is unchanged: write as though a human could read it, because one might. If you would not want a stranger at the company to see a message, that message does not belong in a companion app. Our note on what companion apps actually collect goes through the specific questions to ask of any privacy policy before you trust it.

Does Character AI report you to the police?

There is no public evidence of routine automated reporting of users to law enforcement. What actually happens on flagged conversations is safety classification and crisis resources being surfaced in the app. Separately, like any US company, a platform can disclose user data in response to valid legal process or a genuine emergency.

One thing that is now law rather than policy: SB 243 requires operators in scope to maintain and publish a protocol for handling users who express suicidal ideation, including referral to crisis services, and to report annually to California's Office of Suicide Prevention. So the crisis response you see is a statutory duty in California, not a courtesy.

What the law now requires of any companion app

SB 243 is worth knowing about as a buyer, because it gives you a floor to hold vendors to. It defines a companion chatbot in statute and imposes duties: disclosing that the user is talking to AI, publishing that self-harm protocol, and for known minors adding disclosure, three-hour break reminders, and blocking sexual content. It carries a private right of action with damages of the greater of actual damages or $1,000 per violation, plus fees.

New York and Maine have moved on similar ground. The practical effect for a US buyer is that "we take safety seriously" is no longer the whole answer. You can ask where the published protocol is.

How to use any companion app safely

None of this is unique to one vendor, so the checklist travels. Keep real identifiers out of chat: legal name, employer, address, anything you would not put in a forum post. Check whether the privacy policy says plainly if chats train models, and whether you can opt out. Look for a delete control that removes conversation history rather than just deactivating a login. Assume the character's behavior can change with a model update, and do not build a routine you cannot afford to lose. Export or save anything that matters to you.

And decide up front what you want the thing for. A platform built around thousands of user-made characters is optimized for variety and discovery. If what you want instead is one consistent companion who remembers you, that is a different product, and the alternatives to Character AI are worth comparing on memory and content policy rather than on library size.

So, is Character AI safe to use in 2026?

For an adult who treats it as a product rather than a confidant, yes. It is more regulated, more age-gated and more scrutinized than any competitor, largely because it was forced to be. For under-18 users the open-ended chat product no longer exists, which is the correct answer to the question parents were asking. The remaining risks are the category's, not this platform's alone: intimate data held by a company, and a personality that can change when the model does.

Aroused is adults-only by design, never explicit by permanent policy, and funded by subscriptions rather than engagement, so there is no incentive to keep you talking longer than you meant to. You can try the demo in your browser before creating anything.