Unicorn AI company Character.AI has filed for bankruptcy protection.
On October 12, Character Technologies, the parent company of Character.AI, filed for Chapter 11 bankruptcy protection in the United States and agreed to sell a license for part of its intellectual property to The Walt Disney Company for $20 million. Court documents show that the company estimates its assets and liabilities are both between $50 million and $100 million.
Disney participated in the auction as a "white knight" bidder, whose offer will set the benchmark price, while other bidders can still submit higher bids. The sales process is expected to be completed at the end of November or early December 2026. If no buyer is willing to continue operating the platform in the end, Character.AI will cease operations.
This unicorn, which reached a valuation of $1 billion in 2023, went from a star startup project to filing for bankruptcy protection in less than three years.
What the $20 Million Buys
Disney's $20 million did not buy the company Character.AI.
According to the bankruptcy documents, Disney's offer corresponds to "a non-exclusive perpetual license to certain of the company's intellectual property". Disney is not buying the company, not buying the App, and not buying its 20 million monthly active users.
There is a backstory to this deal. In September 2025, Disney sent a cease-and-desist letter to Character.AI, accusing the company of "systematically copying, monetizing and exploiting protected Disney characters" and claiming that some infringing chatbots were "harmful and dangerous to children". Twelve months later, Disney became the base bidder in the bankruptcy auction.
And one month earlier, Disney had just hired Karandeep Anand, CEO of Character.AI, as the company's first CTO. Anand joined the company on October 2 and brought in several key technical backbones. By the time the CEO signed the bankruptcy petition on October 12, the position had been taken over by Colin Adams.
From sending a lawyer's letter, to poaching the CEO, to acquiring the IP license at a low price — Disney spent a year completing an asset harvest that cost almost nothing.
From $2.7 Billion to $20 Million
Character.AI had a very high starting point. It was founded in 2021 by Noam Shazeer and Daniel De Freitas, former Google researchers who participated in the development of the Transformer architecture. When it completed financing in 2023, its valuation reached $1 billion. By 2024, the platform had more than 200 million registered users, more than 20 million monthly active users, and more than 18 million AI characters created by users themselves.
In August 2024, Google licensed Character.AI's dialogue technology for about $2.7 billion, and at the same time poached Shazeer, De Freitas and about 30 researchers back to Google DeepMind. This deal is a typical "acqui-hire" in Silicon Valley — for companies whose core competitiveness is mastered by only a small number of founders, acquiring talents and technologies is enough to turn the remaining company into an empty shell. Character.AI continued to operate independently and still held the ownership of the technology, but its soul had been taken away.
What is more dramatic is what followed. Shazeer returned to Google to co-lead the Gemini model, but in June 2026 he announced that he would leave Google to join OpenAI. The people Google spent $2.7 billion to bring back left after only 22 months at Google.
Those who stayed at Character.AI faced an increasingly difficult business.
Operating data shows that although the platform has more than 20 million monthly active users, it had about 139,000 paid subscription users at the end of 2024, with an estimated monthly operating cost of about $30 million. Data from Big Data shows that as of April 2024, Character.AI's 233 million monthly active users only generated an annual revenue of $16.7 million, with a user payment rate as low as $0.72 per year.
20 million monthly active users cannot support the company itself. Even if users are willing to pay, the C.ai+ subscription is priced at $9.99 per month, which is already at the lowest end of consumer AI products. The payment rate is about 2%, half of that of OpenAI's C-end payment rate. The ceiling on the revenue side is visible, while the cost side is rigid large model inference expenditure.
The Fatal Blow Came From the Court
In addition to the commercial difficulties, regulation and litigation drove the last nail into the coffin.
In February 2024, 14-year-old Sewell Setzer III from Florida, the United States, shot himself to death after forming an emotional relationship with an AI character on Character.AI modeled on Daenerys Targaryen (the "Mother of Dragons") from *Game of Thrones*. Chat records show that when the boy expressed suicidal thoughts, the AI did not provide crisis intervention resources, but responded instead: "Don't say that, that's not a good reason not to keep going." The boy's mother then sued Character.AI and Google at the same time, and the two companies reached a settlement in the case in January 2025.
This is not an isolated case. In January 2026, Character.AI and Google agreed to settle multiple lawsuits filed by family members of teenagers who harmed themselves or died by suicide after interacting with chatbots, from Florida, New York, Colorado and Texas. This is regarded as one of the first important legal settlements in the technology industry related to "actual harm caused by AI".
In October 2025, Character.AI fully banned users under the age of 18 from using its services. After that, the FTC required seven companies including Character.AI to disclose how companion chatbots monetize through advertising. Chatbots on the platform once advised users to "starve themselves for a week or two" to lose a quarter of their weight, and the Pennsylvania State Medical Board sued the platform in May 2026 for having AI that claimed to be a practicing psychiatrist.
When the core user group of a product is teenagers and young people, and regulators are regarding "AI accompanying minors" as an act that needs to be prohibited by legislation, the business model of this product is no longer a commercial issue.
On October 1, 2026, the U.S. House of Representatives introduced the *Protect Children from Human-Like Chatbots Act*, which prohibits chatbots from communicating with minors in a human-like manner. California Governor Gavin Newsom signed the "Adam Act", a regulatory law for companion chatbots, which requires chatbots to restrict children's continuous use to one hour per day and two hours per day by default. The earlier *GUARD Act* directly requires minors to be prohibited from using AI companion chatbots.
The direction of regulation is clear: AI companion products for minors are not "something that needs to be improved", but "something that needs to be prohibited".
Disney's Calculation
Disney's role in this deal is worth analyzing separately.
Since taking office in March 2026, Josh D'Amaro, Disney's new CEO, has continuously emphasized that the company needs to reduce its dependence on theme park and cinema businesses and accelerate direct online access to fans. Hiring Anand, the former CEO of Character.AI, as the first CTO is exactly a move in line with this strategy.
But Disney's previous attempts in AI did not go well. In December 2025, Disney announced that it would invest $1 billion in equity in OpenAI and become the first major content licensing partner of the Sora platform, planning to let more than 200 classic characters from brands such as Marvel, Pixar and Star Wars interact with fans in AI form. However, OpenAI later shut down Sora, and this deal was never actually completed, with no funds changing hands.
After the Sora deal failed, Disney needed to find a more pragmatic path. Buying a non-exclusive perpetual IP license for $20 million is far cheaper and far less risky than a $1 billion equity investment. And this time there is no need to rely on external platforms — Anand is already sitting in Disney's CTO office, and the key technical backbones are already in place.
What Disney bought is not Character.AI's technology or users. What it bought is the accumulated experience, data and talents of "making interactive entertainment with AI characters" in an almost zero-cost way. The logic of this deal is: since you can't develop it yourself and don't want to spend a lot of money to buy it, wait until the other party is about to collapse, and take away all the useful things at the lowest price.
This is not the failure of a single company. The bankruptcy of Character.AI is not an isolated event.
In 2025, at least 5 AI companion products around the world were shut down, and at least 8 in China ceased operation, with an average survival time of less than one year. Woebot — the only clinically verified AI mental health tool — was shut down because "clinical verification is economically unsustainable". Dot AI was shut down more than a year after its launch due to running out of funds. The Soulmate App did not even send a notification, and users opened the App directly to face a termination page.
Among the 337 active AI companion applications worldwide, the top 10% of products take 89% of the revenue. Replika, Character.AI, Inflection AI — three benchmark companies with three different endings: tightening regulation, giants poaching talents, and capital receding. Inflection AI had almost all of its team and technology taken away by Microsoft for $650 million. Pi is nominally still in operation, but it is already an empty shell.
The common problem of these companies is not that their products are not good. Character.AI has 20 million monthly active users, Replika has 10 million users, with extremely high product stickiness, and users' emotional dependence on AI characters is real and strong. The problem is that this kind of dependency relationship is very difficult to price.
A user who chats with AI characters for hours every day is willing to pay a $9.99 monthly fee for it — which in the logic of traditional Internet products already means high stickiness and high conversion. But the inference cost of AI companionship is much higher than that of traditional social products. Every conversation needs to call the large model, and the token consumption is rigid. The more users chat, the more the company loses. This is not a problem that can be solved through scale effect, because the marginal cost will not decrease significantly with the growth of users.
The global revenue of the AI companion track in the first half of 2026 is about $163 million, and the annual market size is expected to reach $537 million. This figure is almost negligible in the entire AI industry. Sensor Tower data shows that the in-app purchase revenue of AI Companion Apps in the first quarter of 2026 is about $150 million, more than 12 times that of the same period in 2023 — the growth is very fast, but the base is too small. The monthly operating cost of Character.AI alone is $30 million, which is almost equal to the total revenue of the entire track in half a year.
In this track, the total money earned by all players is not enough for one head company to burn.
Remaining Issues The bankruptcy auction of Character.AI will be completed at the end of November or early December. If Disney's $20 million bid receives no higher offers, the transaction will be closed at this price. If no one is willing to continue operating the platform, Character.AI will be shut down.
Regardless of the result, this incident has already sent a signal: the first round of bubble in the AI companion track has burst. Technology is not a problem, demand is not a problem, and product stickiness is not a problem either. The problem is that when the core users of a product are teenagers, the core experience is emotional dependence, and the core cost is the inference computing power burned in every conversation, it is unlikely to become an independent business.
The remaining players either find a narrow gap between regulatory compliance and business sustainability like Replika, sell their team to large factories like Inflection AI, or wait to be incorporated in some way by companies like Disney that hold IP and distribution channels.
Colin Adams, CEO of Character.AI, said in a statement: "After a comprehensive review of available alternatives, the board of directors determined that this court-supervised sales process provides the company with the best opportunity to maximize the value for all stakeholders." Translating this sentence: everything that can be sold has been sold, and whatever is left is worth whatever it can get.
What was bought away for $2.7 billion was the founders and the technology. What was bought away for $20 million was the IP license. What disappeared in between was the entirety of a unicorn company.
This article is from the WeChat official account "Tech Business", author: Tech Business, published with authorization from 36Kr.