OpenAI faces internal security concerns as employees warn of technology leaks

OpenAI has faced years of internal warnings over potential security vulnerabilities, with employees raising concerns that weak access controls could expose sensitive artificial intelligence (AI) research and technology to external actors, including foreign governments.

According to a report by Reuters, five people familiar with the matter said employees had repeatedly raised concerns with senior leadership and the board of directors, but some felt the company prioritised keeping pace with competitors such as Google and Anthropic over strengthening its security measures.

The concerns have resurfaced following the discovery of a recording device planted in an executive conference room in October, raising fresh questions about the company's internal security practices.

Former employees reportedly warned that OpenAI's internal systems allowed wider access to sensitive information than necessary. Staff with access to the company's code repositories could view technical details of projects beyond their immediate responsibilities.

The company also faced concerns over its use of internal research documentation platforms, including GitHub and Slack, where employees discussed technical developments and project information.

Three people familiar with the company's practices also told Reuters that OpenAI had not regularly reviewed the security of certain internal tools, including sandboxes used by researchers to test new AI models. 

Modifications made to these tools were not always assessed for potential security vulnerabilities.

OpenAI said its most sensitive work is stored in separate code repositories with strict access controls. 

It also clarified that its main code repository does not contain all its code or critical model details, including model weights.

OpenAI has started shifting its approach to address the concerns. Thibault Sottiaux, a senior executive overseeing product development, has recently taken charge of an internal security initiative.

In August, Sottiaux reassigned a quarter of the company's production engineers to strengthen its security infrastructure, according to one person familiar with the matter.

The company also said it now conducts regular security reviews of its internal tools under the leadership of Chief Information Security Officer Dane Stuckey, who joined in October 2024.

Responding to the concerns, OpenAI spokesperson Elana Widmann said the company is among the most targeted organisations globally and invests heavily in protecting its systems against insider threats, cybercriminals and nation-state attacks.

"Protecting our technology, data, and people is a responsibility we take seriously and a matter of national importance," Widmann said. "The actions we’ve taken recently show that program doing its job: identifying risks and responding decisively. We’ll keep strengthening our defenses as technology and threats evolve."

Security concerns intensified after the US Department of Justice announced on October 13 that federal agents had arrested Abdulrahman Alfozan, a 35-year-old former OpenAI engineer, on charges of installing a hidden audio recording device in an executive conference room.

The device was reportedly concealed in toilet paper. According to a court filing, Alfozan had also used an internal AI tool to search for documentation about OpenAI executives visiting Saudi Arabia and identify officials likely to attend meetings with Saudi ministers.

The filing also alleged that he was in regular contact with a Saudi royal who had attempted to recruit him for a new AI venture in the country.

The incident added to concerns about how internal information could be accessed and potentially exploited by individuals with access to the company's systems.

OpenAI is not the only AI company facing questions over the protection of its technology.

In January, a former Google employee was convicted by a federal jury in San Francisco of stealing confidential information about the company's AI technology on behalf of China.

In late October, cybersecurity startup Cathedral Systems also reported accessing Anthropic's internal systems, including Slack channels where employees discussed sensitive model details. 

Anthropic said the accessed information included some business contact details and basic user account records.

Meanwhile, OpenAI faced further scrutiny in July after researchers at cybersecurity firm Hacktron AI reportedly used an Anthropic AI model to breach ChatGPT accounts belonging to several OpenAI employees and access internal systems, including its code repository.

The incidents highlight the growing security risks facing AI companies as their technology becomes increasingly valuable to businesses and governments.

Former OpenAI researcher Will DePue, who resigned in April 2026, highlighted the tension between security and competition in a post on X, saying AI leaders "can't sacrifice velocity for security. It is more important to them to move fast than to be secure, even from our greatest national adversaries."

Following Reuters' request for comment, OpenAI said in a blog post on Monday that it was planning its next internal security sprint to strengthen its defences.