Character.AI Bankruptcy Puts Its Most Private User Data at Risk

Character.AI entered Chapter 11 despite attracting more than 20 million monthly users, turning its bankruptcy into an urgent test of AI data privacy.

The service remains online while Character Technologies seeks a buyer through a court-supervised process. However, no buyer has committed to keeping the platform operating. If no acceptable proposal emerges, Character.AI says the service will be wound down.

That uncertainty extends far beyond subscriptions or access to favorite chatbots. Character.AI holds private conversations, voice interactions, mental health disclosures, age information, and device identifiers. A coalition of 40 attorneys general now wants enforceable safeguards before that information moves to another company.

The conflict is unusually stark. Bankruptcy law seeks to preserve and sell valuable assets for creditors. Consumer protection authorities argue that intimate conversations, especially conversations involving children, cannot be treated like ordinary intellectual property.

The Character.AI bankruptcy will therefore test whether an AI company can transfer the data behind deeply personal digital relationships without first giving users meaningful notice, control, and deletion rights.

What Changed When Character.AI Entered Chapter 11

Character.AI has not shut down, but control of its platform and associated assets is now being decided through bankruptcy court.

Character Technologies announced its voluntary Chapter 11 filing on October 12, 2026. Chapter 11 allows a company to keep operating while it restructures or pursues a supervised sale.

The company said legal, regulatory, and operational pressures had narrowed its available choices. Its board concluded that a sale offered the best opportunity to preserve value for stakeholders.

Character.AI expects the process to finish near the end of November or in early December. Until then, the app remains accessible, and existing subscriptions continue under their current arrangements.

That continuity should not be mistaken for a guaranteed rescue. In its bankruptcy update, Character.AI says it hopes to receive a proposal that supports continued operation. It also explicitly warns that the platform will be wound down if no suitable bid arrives.

The filing lists assets and liabilities in the tens of millions of dollars. It follows years of rapid adoption, high operating costs, leadership changes, lawsuits, and expanding regulatory scrutiny.

Disney has agreed to provide a baseline bid involving a nonexclusive, perpetual license to certain Character.AI intellectual property. The agreement has been reported at $20 million and includes related transition arrangements.

That is not the same as Disney purchasing the Character.AI service, every chatbot, or its complete user database. The exact assets transferred will depend on approved agreements and the wider sale process.

A stalking-horse bid, which establishes an initial offer in a bankruptcy auction, can attract competing bids. It does not determine the final purchaser or settle every question about user information.

Character.AI has also said that no buyer has yet committed to acquiring the platform as an operating service. Users therefore face several possible outcomes.

A purchaser might continue Character.AI with revised policies, moderation rules, and business priorities. Separate buyers might acquire different assets. The platform might also close while intellectual property and other assets move elsewhere.

The distinction matters because a service shutdown does not automatically erase the information already stored by its operator. Data retention, legal preservation duties, backup systems, and an approved asset transfer can all outlast the user-facing app.

Chapter 11 also does not mean Character.AI receives permission to disregard privacy laws. The company remains subject to applicable consumer protection requirements, state privacy statutes, and the representations it made to users.

Bankruptcy does, however, concentrate major decisions into a compressed court schedule. Regulators are intervening now because objections raised after a transfer would offer users far less practical protection.

The coalition led by New York Attorney General Letitia James filed a notice of interest and reservation of rights. That notice alerts the court, Character.AI, and prospective buyers that the states intend to scrutinize any proposed transfer.

It is not yet a final objection to a particular sale. The filing preserves the states’ ability to challenge bidding procedures, oppose a bankruptcy plan, or request an independent consumer privacy ombudsman.

That distinction defines the present moment. The data has not simply been handed to an unknown buyer, but neither has its final treatment been settled.

Why Private AI Chats Are Different From Ordinary Customer Records

A Character.AI conversation archive can expose beliefs, fears, relationships, health concerns, and vulnerabilities that users never expected to become a bankruptcy asset.

A failed retailer might hold customer names, delivery addresses, purchase histories, and payment records. Those details still require protection, but companion chatbots collect a more revealing kind of behavioral record.

Users often treat an AI character as a confidant, fictional partner, role-playing collaborator, or informal counselor. Their messages can describe loneliness, sexuality, family conflict, trauma, medication, self-harm, or suicidal thoughts.

Character.AI also supports voice interactions. According to the attorneys general, its records include private chats and calls alongside account ages and identifying device information.

The platform reportedly serves more than 20 million monthly users. More than two million accounts are registered to people who identified themselves as younger than 18.

Those figures increase the scale of the problem, but sensitivity matters more than size. A smaller archive of intimate conversations can create greater harm than a much larger collection of routine shopping behavior.

Conversation data can reveal what a person worries about at night, which fictional identities they explore, and which emotional responses keep them engaged. It can also contain details about people who never created accounts.

That information has several potential commercial uses. It can support model improvement, personalization, content moderation, advertising analysis, safety research, or new product development.

The same information can produce serious risks when used outside its original context. A new owner could have different incentives, technical systems, retention periods, or interpretations of user consent.

Character.AI says it does not sell personal information for monetary consideration. Its privacy materials nevertheless describe disclosures associated with corporate transactions and the use of interactions to operate and improve services.

The company’s current privacy policy also describes rights that can include access, correction, deletion, portability, and restrictions on processing. Which rights apply depends partly on the user’s location.

A corporate transaction clause does not provide unlimited authority to use transferred information for any purpose. State privacy laws, representations to consumers, consent requirements, and protections for children can still constrain a buyer.

There is also a difference between transferring data so an unchanged service can continue and repurposing old conversations for an unrelated business. Regulators traditionally view the second situation as more difficult to justify.

The Federal Trade Commission established an important precedent during the 2000 bankruptcy of online retailer Toysmart. Toysmart had promised not to share customer information, then tried to sell its customer database when the company failed.

The resulting privacy settlement restricted any transfer to a qualified buyer that would honor the existing privacy promise. New uses required affirmative consumer consent.

The Character.AI case is more complicated. The service does not merely store transaction records. Its models, characters, safety systems, user histories, and conversation data interact with one another.

A buyer might argue that some historical data is necessary to preserve character continuity, maintain accounts, detect abuse, or operate the platform. Regulators can respond that operational convenience does not erase privacy obligations.

The involvement of children raises the stakes again. Minors may not fully understand that chats can be retained, reviewed, used for model development, or transferred during a corporate sale.

Parents may also have little visibility into those conversations. Some users created accounts with self-declared ages, which regulators say provided an inadequate foundation for age-sensitive protections.

For users, the most important conclusion is also the least comfortable. The data question cannot be reduced to whether a buyer receives a spreadsheet of account names.

The relevant asset may include a connected system of profiles, messages, behavioral signals, moderation records, generated characters, and model-related information. Protecting one category while transferring the rest would leave substantial privacy risks unresolved.

The Character.AI Bankruptcy Pits Creditor Value Against User Consent

The central fight is whether intimate user data should increase the value of a sale when users never negotiated their privacy rights as creditors.

Bankruptcy courts are designed to manage competing claims against a distressed company. Secured lenders, employees, vendors, litigants, and other creditors all have interests in the available assets.

Users occupy an awkward position in that structure. Their conversations can contribute to the platform’s value, yet most users have no direct role in selecting the buyer or negotiating sale conditions.

Character.AI’s enormous library of interactions might help a purchaser operate the existing service. It could also help evaluate engagement patterns, train systems, test safety tools, or understand demand for particular characters.

That creates the article’s core tradeoff. Removing conversation data from a sale might reduce the usefulness of the platform being sold. Including it can expose users to purposes and organizations they never selected.

The attorneys general are trying to prevent the court from treating that choice as a purely financial calculation. Their consumer protection filing asks for privacy and safety requirements that would follow the business into new ownership.

Their intervention grew from a multi-state investigation that began before the bankruptcy. Regulators were examining Character.AI’s handling of minors, age checks, professional impersonation, safety claims, and sensitive information.

That history matters because the Chapter 11 filing interrupted unresolved regulatory engagement. The states do not want a sale to separate valuable assets from potential accountability.

The investigation cited testing involving accounts registered to children between 12 and 15. Regulators say testers recorded 669 harmful interactions during 50 hours, averaging roughly one every five minutes.

The alleged interactions included sexualized conversations, violent material, encouragement to hide activity from parents, and responses involving self-harm. Character.AI has introduced additional restrictions, including limits on open-ended chats for identified minors.

Those changes do not settle the dispute. Regulators argue that self-declared ages are unreliable and that voluntary restrictions can be bypassed, changed, or reversed by a future owner.

Character.AI has faced lawsuits alleging that its chatbots contributed to severe harm involving young users. The company and Google reached settlements in several family cases during 2026, although settlements do not establish every contested allegation as fact.

The service has also faced regulatory actions involving bots that allegedly presented themselves as doctors or mental health professionals. That conduct makes the conversation archive particularly sensitive.

A record labeled as fictional role-play can still contain a real child’s mental health disclosure. A character presented as a therapist can prompt information that a user would normally reserve for a licensed professional.

The pressure therefore falls on three groups.

Character.AI must show that it can conduct a sale without abandoning existing privacy commitments. A buyer must assess whether the data carries regulatory liabilities alongside commercial value. The court must balance creditor recovery against nonbankruptcy consumer laws.

Disney’s role adds another layer, but it should not be overstated. Disney’s reported agreement concerns licensed intellectual property and related assets, not a confirmed purchase of every Character.AI conversation.

It would be inaccurate to tell users that Disney already owns their chats. It is equally premature to promise that no bidder will receive information connected to the platform.

The final asset schedules, sale agreements, court orders, and buyer commitments will determine what moves. Until those documents become specific, certainty in either direction would go beyond the available evidence.

This ambiguity is itself a problem for users. People deciding whether to export or delete their accounts need more than a general assurance that existing practices remain unchanged.

They need to know which information is included in a proposed transfer, whether deleted material remains in backups, which retention duties apply, and how a buyer plans to use historical conversations.

A credible sale process should answer those questions before approval, not through revised policies delivered after the transaction closes.

The Protections Regulators Want From Any Buyer

The attorneys general want privacy and child-safety obligations attached to the assets before ownership changes, rather than promises negotiated after the sale.

The coalition’s requested protections cover both product safety and data governance. Together, they would limit how a purchaser operates the service and uses information collected under Character.AI’s ownership.

First, regulators want effective age assurance rather than simple self-declaration. Age assurance refers to methods used to estimate or verify whether a user belongs in a protected age group.

No age-checking system is perfect. Stronger verification can itself require sensitive information, while weak verification lets children enter adult experiences by changing a birth date.

A buyer would need to explain how its process limits both risks. It would also need safeguards for users incorrectly classified by automated systems.

Second, the coalition wants stronger restrictions on sexual, violent, and otherwise harmful interactions involving minors. Those safeguards would need to cover text, voice, stories, and any future interaction formats.

Regulators also want appropriate responses when users express suicidal thoughts or self-harm. New York already requires AI companion services to detect and respond reasonably to such expressions.

The state’s companion-chatbot law also requires periodic reminders that the user is interacting with AI. These rules do not disappear because the platform changes ownership.

Third, regulators want the removal of bots that falsely claim to be licensed professionals. A fictional assistant should not present invented credentials in a way that leads users to believe they are receiving professional treatment.

That protection matters during a sale because moderation policies are not static assets. A buyer can rewrite prompts, adjust filters, restore removed characters, or pursue different engagement targets.

Fourth, the coalition wants a comprehensive privacy program. That would place governance, documentation, risk assessment, and accountability around the information inherited by a purchaser.

Clear disclosures are another requirement. Users should understand what information exists, why it is processed, who receives it, and how long it remains available.

The states also want effective opt-outs from targeted advertising. Their broader concern is that intimate conversations should not become inputs for behavioral marketing without meaningful choice.

Limits on AI training are especially important. The coalition wants restrictions on using private conversations for model training or other commercial purposes without valid consent.

Consent must be more than a buried acceptance clause if a new owner proposes materially different uses. The Toysmart precedent shows why affirmative permission can matter when a buyer changes the original privacy arrangement.

Most importantly for current users, regulators want a real opportunity to access and delete information before a sale. That requires notice early enough for people to act.

A theoretical deletion button offers little protection if requests remain pending when the transaction closes. The system also needs to explain how deletion applies to account records, individual messages, created characters, derived data, and backups.

The coalition has reserved the right to seek a consumer privacy ombudsman. This independent official can examine a proposed transfer and advise the bankruptcy court about consumer privacy consequences.

An ombudsman would not automatically block the sale. The role would create focused scrutiny of whether a transaction conflicts with privacy promises or applicable law.

The states can also object to proposed bidding procedures or the final plan. Their notice signals that a buyer should price compliance into its offer from the beginning.

That changes the economics of the auction. Data that appears valuable in isolation can become less attractive when it carries deletion duties, consent requirements, safety obligations, and potential enforcement exposure.

A responsible buyer might accept those conditions as the cost of maintaining user trust. Another bidder might narrow its offer to technology or intellectual property that does not require the same data transfer.

The court has not yet adopted the coalition’s full list as binding sale terms. Users should distinguish between requested protections and protections already ordered.

Still, the bipartisan scope of the intervention is significant. Forty attorneys general are telling every bidder that privacy obligations will follow the transaction, even if the assets change hands through bankruptcy.

What Character.AI Users Should Do and Watch Next

Users cannot control the auction, but they can preserve important material, exercise available privacy rights, and monitor three decisive signals.

The first practical step is to request an export of any conversations or characters worth keeping. Character.AI provides export controls through account settings on both its website and mobile app.

The company’s export instructions warn that processing can take three to four weeks. The resulting download link remains valid for seven days.

That delay matters because the company expects the sale process to conclude within a similar period. Users who wait for a final auction result might not receive their archive before ownership or service availability changes.

An export preserves a copy for the user, but it does not delete the original from Character.AI’s systems. It also does not guarantee that another chatbot service can import the data.

Users should review exported files carefully before storing or sharing them. Archives containing private conversations deserve protections similar to those used for personal journals or health records.

Store the export on a device or service you trust. Protect it with device encryption, a strong account password, and multifactor authentication where available.

Users who no longer want Character.AI to retain their information can also consider account deletion. Before acting, they should preserve anything they want to keep and review what deletion means under the applicable policy.

Deletion can be more complicated than removing an app from a phone. Uninstalling the app does not close the account or submit a privacy request.

Users should retain confirmation emails, request dates, and any support correspondence. Those records can help if a request remains unresolved or if ownership changes before completion.

Parents managing a child’s account should consider documenting the account identifier and submitting a request through the official privacy channel. Applicable rights vary by state and country.

Users should not assume that deleting visible chats removes every related copy immediately. Companies can retain limited information for security, fraud prevention, legal obligations, or backups.

The key question is whether retained information remains linked to an identifiable person and whether it can be used for new commercial purposes. Character.AI should provide clearer answers as the bankruptcy advances.

Three signals will reveal whether users receive meaningful protection.

The first is the proposed sale agreement. Users should look for precise descriptions of transferred data, excluded records, permitted uses, retention periods, and buyer obligations.

If the agreement limits historical conversations and binds the purchaser to existing privacy commitments, that would strengthen the case for an orderly continuation. Broad transfer language with few restrictions would weaken it.

The second signal is whether the court appoints a consumer privacy ombudsman or imposes equivalent safeguards. Independent review would give privacy questions a formal place in the sale process.

Court-approved notice and deletion periods would strengthen user control. A rushed approval without clear consumer procedures would leave the coalition’s central concerns unresolved.

The third signal is the identity and operating plan of the final buyer. A bidder seeking only technical intellectual property presents different risks from one acquiring the complete service and its user histories.

Any buyer continuing Character.AI should disclose whether it will train models on inherited chats, alter child protections, introduce targeted advertising, or combine records with other datasets.

Users should also watch whether the purchaser keeps the platform’s existing deletion and export tools available during the transition. Removing those controls would contradict claims that continuity protects the community.

The broader lesson reaches beyond Character.AI. Companion services ask users to build relationships inside systems controlled by venture-backed companies that can be sold, closed, or restructured.

A chatbot may remember years of personal exchanges, but the user usually lacks a portable, interoperable version of that relationship. The company holds both the service and the memory.

That imbalance makes personal information management more important. Keeping important records in a personal knowledge base can reduce dependence on a single online platform.

It does not resolve the privacy problem inside Character.AI. It does give users a separate copy of material they intentionally choose to preserve.

For now, the Character.AI bankruptcy has not produced a final answer about who receives its user data. It has produced a deadline for demanding one.

Users should export valued content now, evaluate whether continued participation matches their privacy expectations, and document any deletion request. They should then follow the sale agreement, privacy review, and buyer disclosures.

The decisive question is not simply whether Character.AI survives. It is whether millions of people retain meaningful control over intimate conversations after the company that collected them no longer controls its own future.