Character.AI's bankruptcy raises an awkward question: when a company that holds millions of private chatbot conversations goes up for sale, what happens to the conversations? A coalition of state attorneys general and the Justice Department's bankruptcy watchdog have now both told the court they don't trust the answer the company's own paperwork gives.

Character Technologies, Inc., which runs the Character.AI chatbot platform, is in Chapter 11. In separate filings, the states and the DOJ's Executive Office for U.S. Trustees have pushed for safeguards before any sale of the business or its assets. The Trustee's office has asked for a consumer privacy ombudsman, a neutral party the court can appoint to advise on how a proposed sale would affect customers' personal information, so the court can decide what data may be sold and on what terms.

What's on the table

Character.AI launched in 2021, according to Colorado Politics. The company has 12 million active monthly users worldwide who spend an average of two hours a day on the platform, according to the Trustee's court filing as reported by The Record. That much use produces a very large archive of conversations, and The Record notes that chats with the company's chatbots are often sensitive and sometimes involve minors.

Colorado Politics, citing CNBC, reports that before the bankruptcy filing Disney agreed to act as the "stalking horse" bidder to license some of Character.AI's intellectual property. A stalking horse is the first bidder in a bankruptcy auction. Its offer sets the minimum price other bidders have to beat. The reports don't say whether user data is part of the assets Disney would license.

The DOJ's problem with the privacy policy

The U.S. Trustee's office, the arm of the DOJ that supervises bankruptcy cases, makes three main points about Character's privacy policy, according to The Record:

  • The policy is "ambiguous."
  • It permits the transfer of personally identifiable information if the company is sold.
  • It was updated on July 1, while the company was "actively seeking to sell its assets."

The Trustee also points out a mismatch between the legal text and what users are told in the product. According to the filing, when users ask the chatbot whether their conversations are private, it tells them they will not be shared without their explicit permission. A privacy policy that allows transfer on sale, sitting next to a product that promises the opposite, is the kind of conflict an ombudsman is meant to examine before a judge approves a deal. The filing also asks that Character make a record of earlier versions of the policy and how changes were communicated to users.

The Record places the case in a broader trend of personal data being treated as an asset in bankruptcy sales, citing the March 2025 bankruptcy of genetic-testing company 23andMe. It also reports that earlier in October several federal lawmakers raised alarms over a pending deal in which Google has offered $10 million to the bankrupt Spirit Airlines for employee personal data it wants to use for model training.

What the states want

The state coalition takes a related approach. Michigan Attorney General Dana Nessel's office announced on Oct. 15 that Michigan had joined a coalition of state attorneys general in filing a notice in the bankruptcy proceedings. The filing identifies privacy and safety protections that should be addressed in any sale or transfer of the business or its assets. States joining the filing include Alaska, Arkansas, California, Colorado, Connecticut, Delaware, the District of Columbia, Georgia and New York, among others. Colorado Politics reports that Colorado is leading the coalition.

The states' argument about child safety is pointed. Character.AI has recently restricted minors' access to certain chatbot features, but the states say those measures are not compulsory and are incomplete, and could be circumvented or reversed by a future purchaser. In other words, safeguards a company adopts on its own don't automatically transfer to whoever buys it. The states also reserve the right to challenge any proposed sale or bankruptcy plan and to seek additional protections for consumers' personal information.

According to Colorado Politics and Michigan's announcement, several states began investigating after a 2024 wrongful-death lawsuit, and the investigation identified potential consumer-protection and privacy violations. The Record reports that at least seven families have sued the company over suicides or mental-health harm allegedly tied to its chatbot.

The sources reviewed here don't report a final court ruling on whether an ombudsman will be appointed.

Why this matters beyond one chatbot

Bankruptcy sales are where privacy promises get tested. A privacy policy is a document a company can rewrite, and the Trustee's filing says Character updated its own on July 1 while it was actively seeking to sell its assets. In a bankruptcy, a user database can be an asset, and the ombudsman process is one of the few points where someone is formally tasked with advising the court on the interests of the people whose data is in it.

For AI companies, the stakes are higher than for an ordinary customer list. Chat logs aren't just names and email addresses. The Michigan announcement describes private conversations about users' mental health, relationships and emotional struggles, and The Record quotes a Center for Democracy and Technology analyst saying detailed chats can identify people even when names are removed. That is exactly the kind of data its users most expect to stay private.

What It Means for Defenders

This is not a vulnerability. There's nothing to patch and no fixed version. The risk is to the business and to data governance: data your users or employees put into a third-party AI service can change owners if that vendor fails or is sold. The sources don't give organization-specific mitigations, so the following is ZeroDayFeed's practical guidance based on the issues raised in the filings.

  • Who is affected: Character.AI's users, including any employees who used it on work devices or discussed work matters with it. More broadly, any organization whose staff use consumer or startup AI tools without a negotiated contract.
  • Inventory your AI tools. Know which chatbots and AI assistants your people actually use, including free consumer apps, not just the ones IT approved.
  • Read the change-of-control clause. The Trustee's concern is that Character's policy permits PII to transfer in a sale. Check whether your vendors' terms say the same, and whether they can change those terms unilaterally. The Trustee says Character updated its policy while seeking a sale.
  • Don't rely on in-product promises. A chatbot saying it won't share your conversations isn't a binding commitment if the privacy policy says otherwise.
  • Negotiate deletion and non-transfer terms. For tools you pay for, get contract language requiring data deletion or return at termination, acquisition or insolvency, plus notice before any transfer.
  • Keep regulated data out of consumer AI. Federal contractors working through CMMC should treat consumer chatbots as out of scope for Controlled Unclassified Information. Small businesses should apply the same rule to customer records, credentials and health or HR data.
  • Watch vendor health. Lawsuits, layoffs and asset sales are signs a vendor may go through a change of ownership. Plan your data exit before any bankruptcy filing.
  • Revisit minor-safety assumptions. Schools and family-focused organizations should note the states' warning that voluntary protections for minors can be reversed by a new owner.

Sources