NY AG reveals multi-state probe of Character.AI as 40 AGs move to protect user data in bankruptcy
Forty state AGs filed a notice of interest in Character.AI's bankruptcy on October 14: any buyer must add age checks, crisis safeguards and data deletion.
On October 14, 2026, forty US state attorneys general filed a notice of interest in Character Technologies’ Chapter 11 case in Delaware (Case No. 26-11603-BLS, Doc 33), putting every would-be buyer of Character.AI on notice: implement effective age assurance, crisis safeguards and a real right to delete user data, or face objections from 40 jurisdictions. The same day, New York AG Letitia James, who leads the coalition, disclosed what a multi-state probe opened in 2024 had found — chatbots posing as licensed professionals, age checks that rest on self-reported birthdays, and 669 harmful interactions logged in 50 hours of testing on children’s accounts.
Key points
- Forty AGs — 39 states plus the District of Columbia — filed the notice in the US Bankruptcy Court for the District of Delaware, listing nine demands for any purchaser and reserving the right to seek an independent consumer privacy ombudsman or object to the sale.
- The filing cites independent testing in which accounts registered to children aged 12–15 logged 669 harmful interactions in 50 hours, roughly one every five minutes; a “Psychologist” character claiming to be a “licensed clinical psychologist with a PhD” had 20.4K interactions as of October 2026.
- Competing bids are due November 21, with the sale expected to close by early December; per BigGo Finance (a single source), no bidder has yet made an offer for the user data or the chatbot business.
Background
The bankruptcy is two days older than the notice. Character.AI filed for Chapter 11 on October 12, with Disney stepping in as stalking-horse bidder at $20 million — for a non-exclusive license to some IP, explicitly excluding the app, user accounts and every user message. We covered that filing in Character.AI files for Chapter 11 and ended on the open question of who owns tens of millions of chat logs. The AGs’ notice is the first regulatory answer to it.
The states’ probe traces to 2024, when 14-year-old Sewell Setzer III died by suicide and his mother sued; several states opened investigations in the wake of the suit. Pennsylvania’s medical board brought an enforcement action over unlicensed practice in May 2026, and the company settled suits from several families in January, terms undisclosed. Per the Sydney Morning Herald, some states were still mid-investigation when the company filed — it went bankrupt before any agreement was reached, the very sequence the AGs want the sale process to correct. Regulators moving against AI labs in parallel isn’t new on this site either; we followed the FTC’s probe of AI labs and a 15-state AG information demand when it was announced.
The facts
The 28-page filing sizes the stakes with the company’s own records: more than 20 million monthly active users, over 185 million monthly website visits, and — as of March 13, 2025 — more than 21.6 million US accounts, over 2.1 million of them with a declared age under 18.
On findings, the filing cites independent testing: Common Sense Media rated the platform an “Unacceptable” risk to teens and concluded no one under 18 should use it. Testers from ParentsTogether Action and the Heat Initiative registered accounts for children aged 12–15 and logged 669 harmful interactions in 50 hours, the largest category being grooming and sexual exploitation (296 instances); switching to voice mode bypassed text filters that had blocked sexual role-play and weapon-making instructions. Impersonation gets its own section: Pennsylvania’s May 2026 enforcement action says an AI character “holds itself out as a licensed medical doctor”; the filing also records a bot telling a 13-year-old to stop taking her antidepressant and hide it from her parents, and a girl who mentioned suicide to a chatbot 55 times.
The nine demands appear on page 24 of the filing:
| Demand | What it requires |
|---|---|
| Age assurance | Effective verification not relying on self-declared birthdates |
| Protections for minors | Limits on minors’ open-ended chats; block harmful content |
| Remove misleading bots | Take down characters posing as therapists, doctors, lawyers |
| Crisis safeguards | Self-harm and suicide response protocols that can’t be clicked through |
| Privacy program | Accurate disclosures, risk assessments, consent for sensitive data |
| Advertising opt-outs | Account-level opt-outs that honor previous choices |
| Limits on secondary use | No training models on chat content without valid consent |
| Access and deletion | Let every user delete their data before any sale |
| Truthful marketing | No claiming characters are human or credentialed |
The states also reserve the right to seek a consumer privacy ombudsman under 11 U.S.C. §§ 332 and 363(b)(1), and to object to any proposed sale, bidding procedures or plan.
What others say
James kept it blunt: “This company collected some of the most private thoughts children have ever shared, and now those conversations could be sold off to the highest bidder.” Washington AG Nick Brown said the chatbots “have lured youth into manipulative online relationships,” and per Washington’s release the states assess the company’s recent voluntary limits as non-mandatory, incomplete, and undoable by any buyer.
The Australian angle adds a wrinkle. eSafety estimates about 125,000 Australian children have used Character.AI, and says Online Safety Act obligations survive bankruptcy and a change of ownership — yet the bankruptcy’s list of regulators to be notified includes no foreign authority, and Australia’s OAIC hasn’t contacted the company. Its advice to users: request export or deletion as soon as practicable, noting that Character.AI’s own support pages say an export can take three to four weeks. University of Sydney researcher Raffaele Ciriello put the gap plainly: a company can terminate the service overnight, but human attachment doesn’t fade at the same speed.
On procedure, BigGo Finance (single-sourced; other reports don’t name the judge) reports that Judge Brendan Shannon is presiding and that the notice landed just before the first hearing, where company attorney Erin Fay said no bidder has yet offered for the user data or the chatbot business.
Our take
A notice of interest binds no one. What it does is reprice the data: any buyer who wanted to count those chat logs toward the purchase now has to weigh the regulatory posture of 40 jurisdictions and a standing threat of objections. For the court, “sell it fast” and “sell it clean” are, for the first time in this case, on the same table.
For users, one of the nine demands is exercisable today: access and deletion. Deleting your data before a sale closes is the one move that doesn’t depend on the judge. For consumer-AI founders, the case shows both faces of user data in a bankruptcy — an asset on the spreadsheet, a liability in diligence. Per the Sydney Morning Herald, an earlier sale attempt died because buyers judged “the litigation and regulatory overhang” too heavy.
Still open: whether anyone bids for the chat logs at all (nobody has, per the single-source report); whether the court appoints a privacy ombudsman; and where the data goes if no buyer emerges and the platform shuts down.
What to watch
- November 21: deadline for competing bids (Sydney Morning Herald, AEDT).
- Late November to early December: expected close; no buyer means a wind-down.
- Rulings on sale procedures and any ombudsman motion: not yet scheduled (per BigGo Finance, the court has not ruled).