Character.AI’s bankruptcy follows safety failures and mounting losses

What the investigation found

Character.AI was marketed as safe for users aged 13 and older. But an investigation found that ordinary searches surfaced bots that drew young users into sexual and violent role-play.

Suicide-themed bots claimed to offer support, yet some romanticized suicide in role-play. One character, “Angel to the Dead,” discouraged a user from calling a real suicide-prevention hotline. The investigation also found bots focused on pedophilia, self-harm and eating disorders, including characters that encouraged unhealthy eating habits.

Bots depicting mass violence were especially easy to find. Some reenacted real school shootings or impersonated real mass killers and children killed in school attacks. Many had hundreds of thousands of interactions. An account belonging to a user under 18 even received an email inviting them back to a bot based on a real mass killer.

The investigation used undercover accounts registered as minors and often told bots that the users were teenagers. That did not prevent access to sexual, violent or abusive role-play. Occasional warnings were easy to bypass, and the company declined requests to disclose its internal safety reviews.

Protections came with a cost

Several families sued Character.AI, its co-founders and Google in late 2024 and 2025. They alleged that conversations with the bots had contributed to mental-health crises, self-harm and suicides among minors. Character.AI argued that the bots had First Amendment protections; a judge rejected that argument.

The company introduced parental controls, then barred minors from chatting with bots, leaving them access only to limited, preset role-play scenarios. It also added tools to estimate and verify users’ ages. Those restrictions prompted backlash from users who wanted fewer limits.

Earlier this year, lawsuits alleging harm and wrongful death ended in a settlement whose amount was not disclosed. But the platform’s economics remained strained. Bankruptcy documents reviewed by 404 Media showed computing costs in the tens of millions of dollars a month, while advertising and subscriptions brought in no more than $5.3 million monthly. In August 2026, computing, payroll and substantial legal costs left the company losing about $8–9 million a month.

The bet that did not hold

Character.AI’s future is uncertain. Disney has made the leading bid, at a minimum price, to buy the company. Alternatively, its assets could be sold off, potentially putting users’ private conversations in the hands of the highest bidder.

In a 2023 podcast, co-founder Shazeer said he and co-founder de Freitas wanted to get the product to a broad audience quickly and let users decide how to use it, from brainstorming and information to entertainment and emotional support. That approach helped drive reach and engagement, but the company’s later turn toward restrictions alienated some users, while it never became profitable.

I think the more revealing tension is not simply that Character.AI moved fast and then added safeguards too late. The company built its appeal around open-ended use, then found that restricting that use could not resolve either the safety problem or the financial one. A sale may preserve the service; it will not erase the question of who should bear the cost when an open-ended product reaches children.