After a listening device was found in an OpenAI executive meeting room, sources say employees had reported security vulnerabilities for years

After a "bugging" device was found in an OpenAI executive meeting room, people familiar with the matter say employees had been reporting security vulnerabilities for years

2026-11-04 08:00 IT Home - Qingyuan

IT Home, November 4 news: In October of this year, OpenAI discovered a secretly installed recording device in an executive meeting room. Several former employees were not surprised when they learned the news.

According to Reuters, citing five people familiar with the matter, as early as the end of 2022, before ChatGPT was launched, employees had raised questions about OpenAI's security management and reported the issues to company executives and the board of directors. After communicating, some employees believed that management was more concerned with how to catch up with competitors such as Google and Anthropic, and that its attention to security issues was insufficient.

These employees believed that OpenAI's security management was very lax, and that outsiders and even foreign governments could relatively easily obtain the company's sensitive technical information. People familiar with the matter gave an example, saying that any OpenAI employee with code access permissions could view the underlying technical details of a large number of the company's projects.

According to two other people familiar with the matter, OpenAI has now begun adjusting its security strategy, with senior product development executive Thibault Sottiaux leading efforts to strengthen internal system protection.

Employees' years of continuously reporting security vulnerabilities to management also reflect the dilemma facing the AI industry. After ChatGPT was launched at the end of 2022, AI technology gradually became a strategic asset of the United States, and U.S. companies therefore came under greater pressure to prevent their technology from falling into the hands of foreign competitors.

But adding security protection measures may also slow down AI technology research and development, and even add obstacles to the large-scale IPOs that OpenAI and Anthropic are planning.

Former researcher Will Depue, who left OpenAI in April this year, once pointed out that in the face of fierce competition, AI company leaders cannot sacrifice speed for the sake of security.

Regarding the security risks raised by employees and the accusation that management did not take action, OpenAI spokesperson Elana Widmann responded that OpenAI is one of the most frequently attacked institutions in the world, and the company has invested heavily in building a security protection system to guard against attacks launched by insiders, cybercriminals, and nation-state actors.

According to a previous IT Home report, on October 13, the U.S. Department of Justice announced that federal law enforcement officers had arrested a 35-year-old Saudi Arabian citizen, Abdulrahman Alfouzan, accusing him of secretly installing a recording device hidden in toilet paper in an OpenAI executive meeting room. At the time of the incident, Alfouzan was an engineer at OpenAI.

Under OpenAI's past code management approach, as long as employees had code access permissions, they could view the technical details of many projects, including sensitive projects unrelated to their own work.

The reason is that OpenAI stored most of its code in a system known in the industry as a "monorepo." It is large in scale and complex in structure, which may have deterred some potential attackers in the past, but AI agents have already acquired the ability to search and sort through such a huge amount of material.

OpenAI also used to store a large number of research documents on GitHub and Slack. People familiar with the matter said that some employees therefore worried that, as long as someone deliberately searched, key internal company information could still be obtained by others. However, one of the people familiar with the matter and OpenAI both confirmed that the company has imposed access restrictions on sensitive projects.

OpenAI said that many of the company's most sensitive projects are stored in separate code repositories with strictly controlled access. The monorepo contains neither all of OpenAI's code nor the key technical details of the models, such as the numerical parameters that determine how the models operate, namely the model weights.

According to three other people familiar with the company's internal operations, OpenAI also previously had the problem that internal tools lacked regular security reviews, including the "sandbox" environments that researchers use to stress-test new models. Employees often modified tools according to work needs, but the modifications were not necessarily security-tested, and therefore could leave vulnerabilities for external attackers to exploit.

In response, OpenAI said that under the leadership of Chief Information Security Officer Dane Stuckey, who joined the company in October 2024, the company has adopted strong security management measures and conducts regular security reviews of internal tools.

Related reading:

"New progress in the OpenAI 'bugging' case: the Saudi engineer involved seeks bail and denies intent to flee"