Tuesday delivered OpenAI two problems with the same shape: information the company did not intend to give up.
In San Francisco, federal prosecutors charged an OpenAI engineer with planting a hidden recording device in an executive conference room. The Justice Department said Abdulrahman Alfozan, 35, a Saudi national who joined the company’s product team in July 2025, was arrested on Oct. 9 and remains in federal custody. He made his first court appearance Tuesday on a charge of using or attempting to use a device to intercept oral communications.
The device was found on Sept. 30 during a sweep of the room, inserted into a USB port on the back of a television and concealed in toilet paper, according to a criminal complaint. Investigators said it had been installed on or about Sept. 28, that its contents were encrypted, and that it could monitor audio and move data in a way designed to avoid detection. The complaint says the employee planned to place more devices in sensitive areas of the office.
Closed-circuit footage showed Mr. Alfozan near the conference room on several dates before the device was discovered, even though he had no authority to enter it or join the meetings held there. Prosecutors said that before Sept. 28 he used a productivity tool to make repeated inquiries about the company’s building addresses, the capacity of its meeting rooms, its calendars and the locations of what the tool called leadership and executive meetings.
The room mattered because of what was discussed in it. Meetings held there frequently covered confidential technical, commercial and strategic information, including proprietary research, product plans and business-development opportunities, along with nonpublic detail about OpenAI’s technical progress, its priorities, its government engagements and its commercial relationships, the complaint said.
OpenAI said it identified the employee using its own models to analyze the device, then restricted his access and placed him on leave before terminating him on Friday. The company said it referred the matter to the Federal Bureau of Investigation, which made the arrest. In a statement, OpenAI said it has no indications that the individual inappropriately accessed or exfiltrated model weights, source code or customer data.
FBI Director Kash Patel described the case as an alleged effort to capture sensitive communications about the company’s technology. Prosecutors have not alleged that any government directed the activity, and the complaint does not describe a motive. The investigation is continuing.
The second problem arrived from a courtroom. A federal judge ordered OpenAI to hand over records related to its hiring of former Apple employees, granting Apple early access to material in a trade-secrets lawsuit it filed in July. The order covers documents tied to interviews with former Apple staff who applied to OpenAI’s consumer hardware and robotics groups, including interview checklists, guides, manuals, scripts and technical presentations, as well as records showing that interview discussions referred to Apple technology or components.
The judge also allowed Apple to see private messages between two individuals named in the case and approved depositions for three people, among them Tang Tan, OpenAI’s chief hardware officer and a 24-year veteran of Apple’s product design group. OpenAI has been given five days to produce material it has already collected and 14 days for the rest. The judge denied several broader Apple requests, including access to OpenAI’s hardware research.
Apple has said more than 400 of its former employees now work at OpenAI and argued that the movement of people is part of an effort to acquire its confidential information. OpenAI has denied the allegations and said Apple failed to show that any confidential material was taken. Mr. Tan has said he returned his prototypes before leaving and kept only nonconfidential material, including a departure checklist.
Taken together, the two matters describe the same exposure from opposite directions. One is an insider who allegedly went around the company’s own security. The other is a rival using the courts to reach into the company’s files, and finding a judge willing to open a path.
The insider case is the more uncomfortable for an industry that sells trust. AI laboratories hold model weights, unreleased research and customer data, and they have spent two years telling enterprises that their information is safe inside those systems. OpenAI’s own summary of the incident is that its internal tooling worked, that no secret left the building and that the device never produced usable intelligence. That account may well be accurate. It is also the account a company gives when it cannot prove what a microphone heard over six weeks.
The litigation is the slower problem. Discovery in a trade-secrets case runs for quarters, it pulls in hiring records that most technology companies treat as internal, and it invites the question of how OpenAI’s hardware effort was staffed. Apple is not asking for damages in the discovery phase. It is building a factual record, and the record is being assembled from OpenAI’s own paperwork.
Both cases will take months to resolve, and neither has produced a finding of wrongdoing. What is already visible is a change in the ambient risk around the company. Hiring has become a legal exposure, and a conference room has become a place where a television can be a security perimeter.
