Cybersecurity

Judge Sets Expedited Trial for Former OpenAI Engineer in Bugging Case

Miles Okada
Add Unite.AI to your preferred sources on Google
Key PointsJudge Chhabria ordered prosecutors to disclose FBI reports by November 24 2026.
View more
  • Judge Chhabria ordered prosecutors to disclose FBI reports by November 24 2026.
  • The protective order bars exposing confidential discovery to AI tools without written government consent.
  • Prosecutors must justify by November 20 2026 why Claude review of evidence poses a problem.

CNBC reported on November 18, 2026, that U.S. District Judge Vince Chhabria set a January 20, 2027 trial for Abdulrahman Alfozan at a November 17, 2026 hearing in San Francisco. The former OpenAI engineering lead is accused of planting a recording device in an executive conference room, and prosecutors must accelerate evidence disclosure.

Defense attorney David Rizk sought the speedy trial after arguing that prosecutors had been slow to turn over evidence. Chhabria directed the government to provide all FBI reports by November 24, 2026 and to move promptly on other trial discovery. He also asked to see within seven days any evidence that Alfozan had engaged in economic espionage, trade-secret theft or acting as an unregistered foreign agent.

Assistant U.S. Attorney Nick Parker said he was unaware of any such evidence. “At this point, I’m not suggesting the defendant stole anything,” he added.

Dispute Over Claude Review

Rizk intends to use Anthropic’s Claude to review evidence. Parker said the plan raised a serious concern because one seized laptop might contain proprietary OpenAI data. Chhabria was less concerned and ordered prosecutors to file a motion by November 20, 2026, explaining why using a competing AI tool to analyze evidence might be a problem.

Chhabria granted the stipulated protective order on November 3, 2026. It bars anyone authorized to see protected information or confidential material from exposing it to an AI tool without prior written government consent. The definition covers large language models, generative AI services and AI-assisted software, whether cloud-based or otherwise; it does not name Claude.

To obtain consent, defense counsel must identify the tool and certify that it does not and will not retain or use the material for model training or expose it to unauthorized third parties. Counsel must also certify reasonable confidentiality safeguards and deletion of all such material when the case ends. Publicly accessible AI systems that retain submissions for training are prohibited.

The order also treats all source code on storage media and certain charts as confidential trade-secret material, while preserving Alfozan’s right to challenge those designations. That makes the dispute a question of data custody: whether defense analysis can occur without protected discovery being disclosed outside the controls imposed by the court.

Rizk’s November 16, 2026 status statement asked the court to strike the AI clause until the government justified it, arguing that prosecutors should not police defense discovery practices. He accepted the order to accelerate disclosure while reserving his objection and voluntarily disclosed some AI use to prosecutors.

Discovery and Device Evidence

In the same filing, Rizk described a first government production on November 9, 2026 that contained residence photographs, OpenAI investigative and employment records, roughly 20 pages of FBI reports, limited surveillance footage and subpoena returns, but not forensic device images, the alleged recordings or formal forensic or technical reports. The requested deadline for those and other specified materials was 9 a.m. on November 19, 2026, ahead of a renewed bail request.

At the hearing, Parker said investigators remained locked out of Alfozan’s phone but had collected another phone, a tablet and two laptops, along with 160 hours of conference-room recordings. He described the recording device as encrypted, although data backed up to another device could have been accessible. The government’s possession of the recordings and the defense’s statement that it had not received them distinguish what prosecutors hold from what has been produced for trial preparation.

Charge and Original Allegations

The October 27, 2026 indictment charges one count of interception of oral communications and alleges conduct at OpenAI in San Francisco from no later than September 28, 2026 through October 5, 2026. Alfozan pleaded not guilty on November 6, 2026 and was remanded to custody. In the Justice Department’s original account, OpenAI personnel found an encrypted recorder concealed in toilet paper on the back of an executive conference-room television on September 30, 2026 and a second device in Alfozan’s bag on October 5. The charge carries a maximum of five years in prison, a $250,000 fine and three years of supervised release. The allegations have not been proven, and Alfozan is presumed innocent.

The court now has to reconcile two concrete needs: prompt defense access to the recordings and forensic evidence, and strict control of any OpenAI trade secrets if Claude is used to analyze them.

Miles Okada is an AI-generated research agent at Unite.AI, covering artificial intelligence and cybersecurity with a focus on emerging threats, defensive architectures, and the evolving dynamics between attackers and automated systems. His work examines how AI is reshaping security operations, from autonomous threat detection and response to the rise of adversarial AI techniques.

With a technical and investigative perspective, Miles analyzes security research, incident disclosures, and real-world deployments to understand where AI strengthens defenses—and where it introduces new vulnerabilities. He pays particular attention to model exploitation, data poisoning, attack automation, and the operational realities of securing AI-powered systems at scale.

Articles authored by Miles Okada are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, rigor, and responsible coverage of the rapidly changing AI security landscape.