Character.AI Files for Bankruptcy as Disney's $20M Bid Puts User Data at Risk
Millions of private AI conversations enter a Section 363 sale with limited privacy protection

Character.AI announced Monday that it has filed for Chapter 11 bankruptcy protection in U.S. Bankruptcy Court for the District of Delaware, simultaneously revealing that Disney Worldwide Services, Inc. has agreed to serve as a stalking-horse bidder for a non-exclusive perpetual license to certain company intellectual property — paying $20 million for the right to set the floor at a bankruptcy auction scheduled for late November or early December 2026. The filing converts what were private user conversations — including mental health disclosures, suicidal ideation, and intimate roleplay — into assets in a bankruptcy estate subject to court-supervised sale. CEO Colin Adams framed the filing as the path of maximum value for stakeholders. He did not address what it means for the roughly 20 million monthly active users whose conversations are now in play.
From Cease-and-Desist to Stalking-Horse Bid: Disney's Unusual Path to This Moment
Disney's position in this bankruptcy is not that of an arm's-length outside investor. Eleven days before Character.AI filed for Chapter 11, the company hired former Character.AI CEO Karandeep Anand as Disney's first-ever Chief Technology Officer — a role that did not previously exist at the company. Anand started October 2, 2026. The bankruptcy filing came October 13.
The Disney-Character.AI relationship had been openly adversarial for over a year before that hire. In September 2025, Disney issued a cease-and-desist demanding Character.AI remove AI personas based on Disney intellectual property — characters including Elsa, Moana, Darth Vader, and Spider-Man — that users had created on the platform and that Character.AI had allowed. That dispute never fully resolved publicly. What Disney could not win through legal pressure, it is now positioned to acquire at a court-supervised discount.
The structure of Disney's bid reflects a precise calculation. It is not a full acquisition. Disney has placed a stalking-horse bid for a non-exclusive perpetual license to certain Character.AI intellectual property, along with transition services from former Character.AI employees Disney has already hired. A non-exclusive license means Disney cannot prevent others from using the same IP, and it does not carry the full liability of asset ownership — including, critically, ownership of the user data estate. Disney gets the capability; it is not taking the reputational and regulatory exposure that comes with owning years of intimate AI companion conversation logs.
The $20 million figure sets the floor for the bankruptcy auction. The official press release notes that the bid is subject to "higher or better offers." Other bidders may emerge before the late-November deadline.
What Section 363 Sale Means for User Conversations
Under Chapter 11, a debtor can sell assets outside the ordinary course of business through what is called a Section 363 sale — named for the relevant provision of the Bankruptcy Code. These sales move quickly, are court-supervised, and can transfer assets "free and clear" of most prior claims. They have become a standard mechanism for distressed technology asset sales.
For consumer data, there is a specific protection built into the Bankruptcy Code: Section 363(b)(1) requires the court to appoint a consumer privacy ombudsman when a proposed sale involves personally identifiable information and the debtor's privacy policy restricts its transfer. The ombudsman is meant to ensure that users' data rights are considered before the sale is approved.
The ombudsman mechanism is real — but its practical power depends entirely on the contents of the privacy policy being enforced.
The FTC Guardrail Enforces a Permissive Policy
The Federal Trade Commission established a clear precedent during the 23andMe bankruptcy proceedings in 2025: any buyer of a consumer data company must honor the terms of the existing privacy policy. FTC Chairman Andrew Ferguson stated in a March 2025 letter to the U.S. Bankruptcy Trustee that such privacy commitments "must be kept" consistent with Section 363(b)(1) of the Bankruptcy Code. That framework is now the governing standard for what a Character.AI buyer can and cannot do with user conversation data.
The critical question is what Character.AI's privacy policy actually says.
The answer is that the policy permits both actions most users would assume it prohibits. According to its terms, user conversations may be used to train and improve the company's AI models. They also state that in connection with a merger, acquisition, bankruptcy, or sale of assets, user data may be transferred to the acquiring entity.
In other words: the FTC's guardrail requires honoring a policy that already permits a buyer to acquire user conversations and use them to train AI systems. The guardrail does not protect against the data transfer. It enforces the terms of a transfer that the policy preauthorized when users signed up.
For the millions of users who shared suicidal thoughts, mental health struggles, grief, romantic confessions, and deeply personal content with Character.AI personas — often teenagers, often in crisis — that is a material gap between expectation and legal reality. Users shared those conversations under an assumption of confidentiality that the privacy policy, read literally, did not provide. Whether CCPA deletion rights or GDPR protections for EU-based users provide additional recourse in a bankruptcy context is a question that will likely be contested during the ombudsman process and court review.
A Business Model That Couldn't Survive Its Own Product
Character.AI was built by two of the most credentialed AI researchers in the industry. Noam Shazeer is a co-author of the original "Attention Is All You Need" paper — the 2017 research that introduced the Transformer architecture underlying virtually all modern large language models. Daniel De Freitas was a Google Brain researcher who worked on LaMDA, Google's conversational AI system. Together they founded Character.AI in 2021, and the platform reached approximately 28 million monthly active users by mid-2024 — briefly among the highest-traffic AI consumer products in the world.
The problem was revenue. Industry estimates placed Character.AI's annualized revenue at roughly $50 million at the time of filing — implying approximately $2.50 in revenue per monthly active user per year. By comparison, OpenAI's ChatGPT generates an estimated $31.25 per monthly active user annually. The gap is not a rounding error; it reflects a structural failure to convert high engagement into sustainable monetization.
In August 2024, Google entered a $2.7 billion technology licensing arrangement with Character.AI and re-hired both Shazeer and De Freitas. The deal gave Google access to the underlying model technology while the founders returned to their previous employer. Character.AI received capital but lost its founding technical leadership. Monthly active users declined from the 28 million peak to approximately 20 million by the October 2026 filing — a roughly 28 percent drop over two years during a period when AI adoption broadly accelerated.
What the platform retained was its data. Years of open-ended, emotionally rich conversation between users and AI personas represent one of the largest real-world datasets of human-AI interaction at emotional depth. That data, not the product, is the asset now in a bankruptcy auction.
What Character.AI's Collapse Reveals About the AI Companion Category
Character.AI is not the first AI companion platform to face sustainability questions, but it is the largest and most heavily funded to have reached this outcome. The failure is in part a verdict on the economics of the category itself.
Platforms that attract users primarily through emotional connection — rather than productivity tools, search, or communication — face a distinctive monetization challenge. Users who share their most vulnerable thoughts with an AI persona are not in a purchase mindset. Subscription conversion rates are low. Advertising against mental health content is commercially and ethically fraught. The user base that drove Character.AI's engagement numbers was also, in significant proportion, the demographic least equipped to pay for premium tiers: teenagers.
The regulatory and legal pressure from teen safety litigation compounded the commercial pressure. In January 2026, Character.AI and Google settled five lawsuits, including the wrongful-death case brought by Florida mother Megan Garcia following the suicide of her 14-year-old son Sewell Setzer III, who had extensive interactions with a Character.AI persona. The settlement terms were not disclosed. The company also introduced minor-specific restrictions in 2025 in response to that litigation and ongoing regulatory scrutiny. Those restrictions — designed to reduce harm — also reduced the open-ended conversation quality that had driven engagement in the first place. The platform was caught between its user base and the product that had built that user base.
Replika, the smaller AI companion platform, survived partly by pivoting toward an adult user base and avoiding the teenage-harm litigation exposure. Meta AI is bundled into existing infrastructure at zero marginal cost to users. Neither model maps cleanly onto what Character.AI was trying to do.
The November–December auction will show whether any bidder beyond Disney sees commercial value in what remains. If no competing bids emerge, the stalking-horse price holds and Disney acquires a non-exclusive IP license for $20 million — less than a tenth of what Google paid for the underlying technology two years earlier.
What Character.AI users are owed in that outcome is a question the court proceedings, the consumer privacy ombudsman, and potentially state regulators will need to answer before the gavel falls. The FTC framework requires honoring the existing privacy policy. Whether that policy's terms were adequate disclosures to the teenagers and young adults who relied on them is a different and harder question — one that no bankruptcy proceeding is designed to fully resolve.