Character.ai Files for Bankruptcy as Court-Supervised Sale Determines Platform’s Future

01Character.ai Files for Bankruptcy and Seeks a Buyer, With Shutdown Looming if No Suitable Bid Emerges

Character.ai has voluntarily entered Chapter 11 bankruptcy to pursue a court-supervised sale of the company, while keeping its AI character-interaction platform available to users. The company said it hopes a buyer will enable the service to continue operating, but warned that the platform will be wound down if the sale process does not produce an actionable proposal.

Character.ai is using Chapter 11 to pursue a court-supervised sale of all or substantially all of its assets while the platform remains operating. Character.ai said it took the step after facing significant legal, regulatory and operational headwinds over the past year. Its board reviewed the available alternatives and concluded that a supervised sale offered the best opportunity to maximize value for stakeholders and preserve service for users.

The company plans to sell all or substantially all of its assets through the bankruptcy case. It currently expects the process to conclude by the end of November or, potentially, in early December 2026. Character.ai did not identify prospective buyers, disclose any offers or provide details about its financial position. It also did not explain how much each category of pressure contributed to the bankruptcy decision.

For now, users can continue accessing all Character.ai services through the app and website. The company said its focus during the process remains finding a path forward for the platform and its community while maintaining what it described as a safe and engaging user experience.

Paid subscriptions also remain active, and billing is unchanged for the time being. Character.ai said it intends to notify subscribers in advance if the service changes in a way that affects paid features. That leaves current subscriptions intact during the sale effort, although their longer-term status depends on the platform’s future.

The company said its privacy and security practices remain unchanged and that user information continues to be handled under its existing privacy policy. Users may still delete their accounts or export their data through the Character.ai app or website, giving them a way to retrieve their information while the bankruptcy process proceeds. Court filings and additional case information are available through the company’s claims and noticing agent, Stretto.

02Google Launches Gemini 4 Argon, With API Pricing Starting at $2 per Million Input Tokens

Google has launched Gemini 4 Argon, an AI model designed for complex, multi-step work across coding, finance, law, cybersecurity and science. The company says Argon emphasizes depth and precision and ranks first on the Vals Index, a benchmark covering capabilities in economically important fields, though no independent testing was provided to verify that result.

Argon is now available globally to developers and enterprises through APIs in Google AI Studio and the Gemini platform. Google AI Ultra subscribers can use it in the company’s Antigravity development environment, while Ultra and Pro subscribers can access it through the Gemini app and Search’s AI Mode.

The introductory API price is $2 per million input tokens and $10 per million output tokens. Tokens are the units into which AI systems divide text and other data for processing. Cached input tokens receive a 95% discount. Google did not say how long the introductory prices would remain available.

Google presented internal engineering work as evidence of the model’s practical capabilities. It said teams using Argon in Antigravity increased agentic code submissions—code changes produced through AI-directed workflows—by more than 35%, while reducing code rollbacks by nearly one-third. Those figures describe Google’s own use and are not a promise of equivalent results for customers.

In one project, Argon helped Chrome engineers separate core webpage handling from a legacy folder containing roughly 6 million lines within Chromium’s 40 million-line codebase. According to Google, the model helped complete an architectural refactor spanning hundreds of files in weeks instead of months. The change is intended to let live webpages move between tabs, side panels and floating windows without reloading, while eliminating a category of browser crashes.

Google also said Argon worked autonomously overnight on its open-source Agent Development Kit for Python. The resulting optimizations, now live in the project, reduced in-memory session preparation time by 40% during long sessions and accelerated cold imports by 21%.

Other submitted or upstreamed work included changes that Google says made the xTB quantum-chemistry package up to 2.3 times faster at predicting molecular structures, as well as optimizations that increased Gemma 4 E2B inference speed from about 37,900 to more than 108,000 tokens per second on an A100.

03OpenAI Engineer Charged With Hiding Recording Device in Executive Conference Room; Company Says It Found No Model or Customer Data Theft

A 35-year-old OpenAI engineer has been arrested and charged with using or attempting to use a device to intercept oral communications after employees found a concealed recorder in an executive conference room at the company’s San Francisco headquarters. The room regularly hosted discussions involving proprietary research, product plans, government engagements and other confidential matters, according to the federal complaint.

OpenAI personnel discovered the device during a technical surveillance sweep on Sept. 30. The complaint says it was plugged into a USB port behind a television, hidden with toilet paper and apparently installed around Sept. 28. Its contents were encrypted, and the source does not establish whether it recorded any meetings, whether the encryption has been broken or whether any material was transmitted.

Prosecutors said surveillance footage placed Abdulrahman Alfozan near the room on several dates even though he was not authorized to enter it or attend meetings there. Alfozan had worked at OpenAI since July 2025 and belonged to its product team, according to the company.

The complaint also alleges that a chatbot account associated with Alfozan’s spouse and sharing identifiers with his account asked how to build, conceal and remotely access a recorder. Prosecutors said the queries covered encrypting an SD card and included the model and unique MAC address of the device later found.

On Oct. 5, Alfozan allegedly entered the conference room and looked behind the television. That day, security staff opened a bag he had left elsewhere and found his wallet, identification and a second similar recorder, prosecutors said.

OpenAI said employees used its models to analyze the device and forensic images before referring the case to the FBI. The company then gradually restricted Alfozan’s access, suspended his physical and network credentials, placed him on leave and secured his work equipment. It coordinated his firing with investigators and terminated him Friday, the day of his arrest.

Alfozan remains in custody after appearing in federal court Tuesday. The charge has not resulted in a conviction. OpenAI said it has found no indication that he improperly accessed or took model weights, source code or customer data, and no evidence that product safety or customer information was compromised.