OpenAI employees raised concerns about the company’s internal security practices for years, including before ChatGPT launched in late 2022, according to five people familiar with the matter. Some staffers shared warnings with senior leaders and board members, and later felt management was prioritizing competition with Google and Anthropic over security, the sources told Reuters.
The concerns included broad access to internal information. Employees with access to OpenAI’s code could see details of projects beyond their own work, the sources said. Much of the code was kept in a single repository, known as a monorepo, while research documents had also been stored on services including GitHub and Slack. OpenAI said sensitive work is held in separate repositories with strict access controls, and that the monorepo does not contain all of its code or critical model details such as model weights.
Three people familiar with OpenAI’s practices also said the company had not regularly reviewed some internal tools, including sandboxes used by researchers to test models. Staff sometimes changed these tools without those modifications being checked for security issues. OpenAI said it now conducts regular reviews of its internal tooling and has a strong security program under chief information security officer Dane Stuckey, who joined in October 2024.
The concerns have drawn renewed attention after the company disclosed in October that a recording device had been planted in an executive conference room. The Justice Department said on October 13 that federal agents arrested OpenAI engineer Abdulrahman Alfozan and charged him with installing a concealed audio recorder there. A court filing also said he used an employee AI tool to find internal documents about executives’ visits to Saudi Arabia and potential meetings with Saudi ministers.
OpenAI spokesperson Elana Widmann said the company is among the world’s most targeted organizations and invests heavily in security against threats from insiders, cybercriminals and nation-states. She said recent actions showed the program identifying risks and responding, and that OpenAI would continue strengthening its defenses.
The company has also begun shifting resources toward security, according to two people familiar with the matter. Senior product executive Thibault Sottiaux recently took charge of an internal effort to secure OpenAI’s systems. One source said that in August he reassigned a quarter of the company’s production engineers to improve security infrastructure. OpenAI said on Monday, after Reuters requested comment, that it was planning its next internal security sprint.
Former researcher Will DePue, who left OpenAI in April, wrote on X that intense competition made AI leaders unwilling to sacrifice speed for security. The debate reflects a wider tension in the field: open research can help attract talent and support innovation, while stronger protections may slow development.
OpenAI is not alone in facing security incidents. In January, a federal jury convicted a former Google employee of stealing confidential information about the company’s AI technology for China. Anthropic said a cybersecurity startup gained access to some Slack channels in an incident reported in late October; the company said the information included some customer and partner contact details and basic user account records.
OpenAI’s security concerns have also followed earlier incidents. In 2023, then-researcher Leopold Aschenbrenner wrote an internal memo warning that China could infiltrate the company, according to three people familiar with the matter. The memo followed an outside hack that exposed details about GPT-4; OpenAI said the attacker accessed only Slack channels available to the whole company. More recently, researchers reported attacks involving ChatGPT accounts and internal systems, including the monorepo. OpenAI said it thanked them for reporting their findings and took steps to reduce the risk of similar attacks.
Sources: Investing.com News.