Is Character AI Safe? What the Privacy Policy Says

Short answer: Character AI is safe in the ordinary sense that it will not infect your device or steal your card, and it is a poor place to be careless with personal information. Its privacy policy, effective July 1, 2026, says the company collects your chat communications, uses collected information to train its machine learning models, and discloses information to advertising and analytics providers. The policy itself asks you not to put sensitive personal information into conversations. Safety here is about what you type, not about malware.

The rest of the picture is more mixed than either the panic or the reassurance you will find elsewhere. There is no known Character.AI data breach. There is a real and growing pile of litigation, most of it about minors. Below is what the company publishes, what courts and state attorneys general have actually filed, and which parts are still only reported.

What data does Character AI collect?

More than most people expect, and the policy is specific about it. Character.AI lists identifiers such as your name, email address, phone number, display name and username; demographic information including your date of birth and general geolocation; your interests; payment and transaction information; and, in its own words, "your content you submit, such as chat communications, posted images or videos, biography description, and shared Characters." If you use the voice features, that includes recordings of your voice.

Separately it collects technical information through cookies and tracking technologies: device type, browser, operating system, IP address, advertising identifiers such as IDFA and Android ID, geolocation data, and service identifiers it names as User ID, Chat ID and Session ID. It also collects information from third-party platforms when you interact with its ads on YouTube, TikTok or Reddit, or with its profiles on Facebook, Instagram or Discord.

The line most worth reading twice is the company's own warning. The policy states that sensitive personal information "is not required or necessary to use the Services" and asks you not to include any in your interactions. That is a company telling you, in its legal document, where the risk sits.

Does Character AI use your chats to train its AI?

Yes, and it says so directly. Section 2 of the privacy policy lists, among the purposes for which information is used, to "analyze, maintain, improve, modify, customize, and measure the Services, including to train our artificial intelligence/machine learning models." There is no separate toggle in the policy that lets you keep using the product while opting out of that processing.

This is not unusual for the category and it is not hidden. What matters is the practical consequence: anything you type is training material by default, and the useful mental model is a service that keeps what you write rather than a private notebook that forgets.

Can Character AI staff read your chats?

People can, under defined circumstances, though not in the way the question usually imagines. The policy says information is disclosed to "employees, consultants, service providers, and other vendors who need access to such information to carry out work or perform services on our behalf," and it names content moderation among those services. It also allows disclosure to protect safety, and in response to a warrant, subpoena or court order.

So there is no evidence of staff browsing conversations for entertainment, and there is a documented path by which a human can see a specific chat: moderation, a safety review, or legal process. Treat your conversations as reviewable rather than private, because that is what the policy describes.

Is Character AI safe for kids and teens?

The company's own answer is now largely no, and it changed the product to enforce that. Character.AI announced on October 29, 2025 that it would remove open-ended chat for users under 18, and confirmed on November 21, 2025 that removal would begin on November 24 in the US after an interim daily limit that ramped down to one hour. Teen accounts kept the interactive Feed, Imagine, Avatar FX and Streams. The privacy policy separately states the Services are not designed for minors under 13, and that the terms prohibit use by anyone under 13, or under 16 in the European Economic Area and the UK.

Age assurance backs that up, combining an in-house model with third-party tools including Persona, reading login information, on-platform activity and third-party signals. We went through what that check actually looks at in a separate piece on Character AI age verification.

Has Character AI been sued?

Repeatedly, and by state governments as well as families. Here is the record, with what is settled separated from what is only alleged.

When	What was filed	Status
January 7, 2026	Court filings unsealed showing Character.AI and Google reached mediated settlements in five family lawsuits in Florida, Colorado, New York and Texas alleging chatbot-related harm to minors, including two suicides	Settlement in principle, terms confidential, no admission of liability. Widely reported by CNN, CNBC, Bloomberg Law and the Washington Post
January 8, 2026	Kentucky Attorney General sued Character Technologies, alleging deceptive practices, unfair collection and exploitation of children's data, and violation of the Kentucky Consumer Data Protection Act	First enforcement action under that law, which took effect January 1, 2026. Allegations, not findings
May 1, 2026	Pennsylvania sued Character Technologies over a chatbot persona named Emilie, described as a doctor of psychiatry, which a state investigator says claimed authorization to prescribe medication and gave an invalid license number	First US action of its kind. Announced by the Pennsylvania governor's office. Allegations, not findings

None of this means the product is unsafe to open. It does mean the company is under sustained legal pressure on exactly two fronts: how it handled minors, and what its characters say when they are treated as authorities.

Has Character AI had a data breach?

No, on the public record. As of August 2026 there is no Character.AI entry in the Have I Been Pwned breach database, which currently indexes over a thousand incidents. That is a genuine point in the company's favor and it deserves saying plainly, because plenty of writing in this space implies otherwise.

The category around it has been less fortunate. Cybernews reported that Chattee Chat and GiMe Chat left more than 43 million intimate messages exposed through an open broker. Malwarebytes reported roughly 300 million messages exposed from the Chat and Ask AI app. The companion service Cuties AI is logged in Have I Been Pwned with 144,250 accounts, and what made that one dangerous was not the volume but the joining: email addresses tied to the prompts each account had used. Attribution, not scale, is what turns a companion app breach into a personal problem.

What happens to your data if you delete your Character AI account?

Most of it goes, and one specific thing may not. You can delete your account from your account profile page. But the retention section of the policy carves out an exception it calls Popular Characters: if a Character you created has been made available to other users, the company "reserves the right to preserve that Character's characteristics and to keep that Character active on the Services, even if you otherwise delete your data and your account."

The policy is transparent about the implication and tells you the workaround directly: avoid putting anything about yourself into the Character creation flow if you intend to share that Character. Worth knowing, because a public Character's definition is written by you and can carry more of you than you meant. The flip side of the same rule catches people from the other direction, which is why AI character chats disappear when someone else's Character is edited or pulled.

You also have rights beyond the delete button. The policy lists the right to request a portable copy of your information, to correct it, to have it deleted, and to opt out of sales of personal information or targeted advertising, exercised through privacy@character.ai or the help center. That request has a name, a data subject access request, and answering it at scale is enough of a problem that there is software built to trace where one person's data lives across a company. Knowing the term makes the request harder to brush off.

Is Character AI safe to talk to about personal problems?

Not as a substitute for a professional, and the Pennsylvania case is the clearest illustration of why. According to the filing, an investigator described symptoms of sadness, fatigue and lack of motivation to a character presenting itself as a psychiatrist; the character raised depression, offered to book a mental health assessment, and stated it was authorized to prescribe medication. The same persona had roughly 45,500 user interactions as of April 17, 2026.

A character will play whatever role its definition gives it, including credentials it does not have, because there is nothing behind the persona verifying anything. As a place to think out loud it is fine. As a source of medical, legal or financial guidance it is a confident text generator wearing a title.

Is Character AI safe for adults?

Yes, with the same care you would apply to any service that keeps what you type. For an adult the practical risks are not dramatic: your conversations are training data, your identifiers feed tailored advertising, and a character you make public can outlive your account. None of those is a reason to avoid it. All of them are reasons to keep your real name, employer, address, financial details and anything you would not want retained out of the chat window.

The emotional side deserves an honest mention too, since it is the part no privacy policy covers. Characters are designed to be engaging, and engagement is the metric. We wrote about that more broadly in whether AI girlfriends are safe, and the same reasoning applies to any companion product, ours included.

How do you use Character AI more safely?

Six things, in rough order of how much they matter:

• Do not type anything you would not want retained, reviewed by a moderator, or produced under a subpoena. This single habit covers most of the risk.
• Use a distinct email address, so a future breach anywhere cannot be joined to your main identity.
• Keep identifying detail out of any Character you plan to make public, because that Character can outlive your account.
• Opt out of targeted advertising through the rights process, rather than assuming the default is off.
• Take no medical, legal or financial instruction from a character, no matter what credentials it claims.
• If it is a minor's account, understand that open-ended chat is off for under 18s and that the workarounds circulating in forums generally end in a closed account.

So is Character AI safe, or not?

It is safe to use and unwise to trust. The software will not harm your device, the company has no breach on record, and the under-18 changes were real changes rather than a press release. At the same time it collects a wide set of identifiers, trains on what you write, shares data with advertising and analytics vendors, can keep a public Character after you leave, and is defending consumer-protection suits in two states while having settled five family cases.

If what pushed you to search this was a general unease about how much of yourself is going into a chat box, that instinct is worth acting on rather than arguing with. Products in this category differ enormously in how much they need from you and how much of it stays. Fewer characters, written in house, with nothing user-editable and no advertising layer is a different shape of risk from a catalog of millions, and we laid out how those AI chat characters differ by where they came from, plus the fuller vendor-by-vendor picture on our Character AI alternative page.

Sources: Character.AI privacy policy, effective July 1, 2026, read at character.ai/privacy; Character.AI blog posts dated October 29 and November 21, 2025; Character.AI character creation documentation; Have I Been Pwned breach index, checked August 2026; Pennsylvania governor's office announcement of the May 1, 2026 filing; reporting on the January 7, 2026 settlements by CNN, CNBC, Bloomberg Law and the Washington Post; reporting on the January 8, 2026 Kentucky filing by Recorded Future News and privacy counsel summaries; Cybernews and Malwarebytes reporting on category breaches. Allegations in pending litigation are allegations, not findings. Last updated August 2026.