Are Character.AI Chats Private? What the Policy Says
TL;DR: No. Character.AI's policy allows conversations to train its models, and its terms let staff access content at any time and for any reason. Perspective AI's private mode keeps history encrypted in your browser, from $14.99/mo, alongside the major model families in one subscription.
Key Takeaways
- Character.AI's privacy policy lists training its machine learning models among the purposes it processes your content for.
- The opt-out stops new content from training generative models. The policy states data may still be used for search, recommendations, and safety classifiers.
- The terms of service reserve a right to access, review, edit and delete content at any time and for any reason. That is a blanket right, not a narrow safety exception.
- No specific retention window appears anywhere in the privacy policy. Any 30-day or 90-day figure circulating in summaries of it is not in the source document.
- Character.AI publishes no end-to-end encryption, enclave, or hardware-attestation claim. Privacy on the platform rests entirely on policy commitments.
Quick Answers
Are Character.AI chats private?
No, not in the sense most people mean. The privacy policy permits conversations to be used to improve and train its models, the terms reserve a broad right of access to content, and nothing on the platform is end-to-end encrypted.
Does Character.AI use my chats to train its AI?
Yes, by its own policy. Training its machine learning models is listed among the purposes the company processes user content for. An opt-out exists for new content and generative models, with stated exceptions.
Can Character.AI staff read your conversations?
The terms of service reserve the right to access, review, screen, edit, modify and delete user content at any time and for any reason. The reasons listed are illustrative rather than limiting.
No. Character.AI's own privacy policy lists training its machine learning models among the purposes it processes your content for, and its terms of service reserve the right to access, review, edit and delete user content "at any time and for any reason". Nothing on the platform is end-to-end encrypted, and no specific retention period appears in the policy at all. If what you actually want is a conversation that does not leave a server copy behind, Perspective AI's private mode keeps chat history encrypted in your browser, for $14.99/mo, and the major frontier and open-weight models sit in that one subscription rather than in a separate plan you would have to switch to.
Every page currently ranking for this question makes similar claims and cites nothing. This one quotes the source documents and dates them, which is the only thing that makes an answer like this worth reading. It sits in reviews because it is an assessment of one product's actual terms rather than a ranking.
Are Character.AI Chats Private? The Short Answer
No. Three things are true simultaneously, all of them from Character.AI's own documents rather than from anyone's interpretation of them:
- Your conversations can train the models. The privacy policy names training its machine learning models among the purposes user content is processed for.
- Staff access is permitted broadly. The terms reserve a right to access and review content "at any time and for any reason", with the listed reasons given as examples.
- Deletion has no stated deadline. The policy commits to keeping information for as long as necessary for the purposes it is processed for, and names no period.
What the Privacy Policy and Terms Actually Say
All quotations below were read from the live documents on 2026-08-18. Both are worth opening yourself, because both are short and neither says what the summaries of them say.
Are your chats used to train the AI?
Yes. The privacy policy's list of processing purposes includes analysing, maintaining and improving the service "including to train our artificial intelligence/machine learning models", and separately names developing new algorithms and machine learning models. This is not an inference from vague language. It is an enumerated purpose.
An opt-out exists, on a dedicated model-training page reachable from account settings. Read what it actually covers: it stops new content from being used to train generative models, and the same page states that data may still be used to improve other parts of the service including search, recommendations, and safety classifiers. Opting out narrows the use, it does not end it.
Can Character.AI staff read your conversations?
By the terms of service, yes, and the grant is wider than the safety framing usually attached to it. The content moderation clause states the company may access, review, screen, edit, modify and delete user content "at any time and for any reason", and offers service development and suspected terms violations as examples of such reasons rather than as the limit of them.
This matters because most coverage of Character.AI describes human review as exception-based, triggered by a report or a classifier. The document does not say that. A blanket right that the company chooses to exercise narrowly in practice is still a blanket right, and a reader deciding what to type is entitled to know which of the two they are relying on.
What happens when you delete a chat or your account?
The policy says information is kept for the time necessary for the purposes it is processed for, unless a shorter period is required by law. That is the whole of it. There is no retention window, no number of days, and no deletion deadline anywhere in the document.
This is worth stating plainly because specific figures circulate widely in write-ups of this question, and none of them appear in the source. If you see a confident retention number attributed to Character.AI, it did not come from the privacy policy.
One concrete carve-out is stated: a Character you created and made available to other users may have its characteristics preserved on the platform even after you delete your data and your account. Your account leaves. The character you built may not.
Why the Question Keeps Being Asked
This query has grown for a reason, and the reason is a documented sequence of events rather than general unease.
Lawsuits and the FTC inquiry
- October 2024. A wrongful-death suit was filed in the Middle District of Florida by the mother of Sewell Setzer III, alleging harm connected to a Character.AI chatbot.
- September 2025. A federal wrongful-death suit was filed in Colorado on behalf of Juliana Peralta against Character Technologies and its founders.
- 11 September 2025. The FTC issued 6(b) orders to seven companies including Character Technologies, seeking information on the child-safety practices of companion chatbots.
- 7 January 2026. A court filing recorded a mediated settlement in principle between Character.AI, Google and the plaintiffs in several of these suits, on confidential terms and with no admission of liability. Whether that settlement has been finalised could not be confirmed as of 2026-08-18.
None of this speaks directly to data handling, and it is the reason data handling started getting asked about. Scrutiny of one kind produces scrutiny of the other.
The under-18 change and Parental Insights
On 29 October 2025 Character.AI announced on its own blog that it would remove the ability for users under 18 to engage in open-ended chat, with the change taking effect no later than 25 November, preceded by a two-hour daily cap that ramped down. The announcement also described new age-assurance measures, in-house and third-party.
Separately, a Parental Insights feature introduced during 2025 sends a parent or guardian a periodic summary for a teen who opts in: time spent, characters interacted with, subscription status. It does not share conversation content. That is a defensible design, and it is also a useful illustration of the distinction this page turns on. Parental Insights withholds chat content from a parent. Neither it nor anything else withholds chat content from the platform.
What Would Actually Make a Chat Private
Privacy claims are worth ranking by how much of each one you can check rather than by how strongly it is worded. Character.AI's claims sit at the bottom of that ladder, not because the company is careless but because it publishes no mechanism, only commitments.
| Question | Character.AI | Perspective AI private mode |
|---|---|---|
| Can conversations train models? | Yes, per the privacy policy, with a partial opt-out | No server copy of the conversation is retained |
| Can staff access content? | Reserved at any time and for any reason, per the terms | History is encrypted in the browser |
| Stated deletion window? | None given in the policy | Nothing is stored server-side to delete |
| What backs the claim? | Policy commitments | A stateless relay and browser-side encryption |
The distinction that matters is mechanism against promise. A policy commitment can be honoured perfectly and you have no way to confirm it from the outside. That is not an accusation, it is a property of the architecture, and it applies to most consumer chat products. AI privacy ranked by what you can verify covers the whole field on that basis, and the Venice AI privacy review works through a single vendor's verifiability claim in the same way this page works through Character.AI's.
If Privacy Is the Reason You Are Asking
Two different readers land here. One wants to keep using Character.AI and wants to know what they are agreeing to, and the answer above is the whole answer: assume the conversation is readable and retained, and type accordingly. The other has decided the answer disqualifies the product and wants somewhere else to go.
For the second reader, Perspective AI's private mode is a different mechanism rather than a stronger promise: chat history is encrypted in the browser and no server copy of the conversation is kept. It is not a companion or roleplay product and does not try to be. The same $14.99/mo Starter subscription covers GPT, Claude, Gemini, and the open-weight families for everything that is not sensitive, and what Character.AI charges is the other half of the comparison if you are weighing the two. If it would be a third or fourth AI subscription, the running total of stacked AI subscriptions and what the alternatives charge are the pages that make that arithmetic explicit.
FAQ
Are Character.AI chats private?
No, not in the sense most people mean. The privacy policy permits conversations to be used to improve and train its models, the terms reserve a broad right of access to content, and nothing on the platform is end-to-end encrypted.
Does Character.AI use my chats to train its AI?
Yes, by its own policy. Training its machine learning models is listed among the purposes the company processes user content for. An opt-out exists for new content and generative models, with stated exceptions.
Can Character.AI staff read your conversations?
The terms of service reserve the right to access, review, screen, edit, modify and delete user content at any time and for any reason. The reasons listed are illustrative rather than limiting.
What happens to my chats if I delete my Character.AI account?
The policy states information is kept as long as necessary for the purposes it is processed for, without naming a period. Characters you made public may be preserved even after you delete your account.
Is Character.AI encrypted?
Not end to end. No encryption architecture, trusted execution environment, or attestation claim appears in the terms or privacy policy as of 2026-08-18. Privacy rests on what the company promises rather than on something you can check.
Is there a more private alternative to Character.AI?
Perspective AI's private mode keeps chat history encrypted in your browser with no server copy, which is a mechanism rather than a promise. It starts at $14.99/mo and includes the major frontier and open-weight models.
A chat that does not leave a copy behind
Character.AI reserves the right to read and to train, in its own terms. Private mode on Perspective AI leaves the transcript encrypted on your device instead, and the same subscription from $14.99/mo covers the mainstream families for everything that is not sensitive.
Try Perspective AI →