OpenAI Staff Warned Leaders About Security Lapses for Years, Reuters Reports

According to the report by Deepa Seetharaman, some staff concluded that management cared more about keeping pace with Google and Anthropic than about security. Workers argued that controls were loose enough that outsiders, including foreign governments, could obtain sensitive technical information with relative ease.

What Employees Flagged

Most of OpenAI's code sits in a single repository, or monorepo. The sources said anyone with code access could see technical details of many projects, including sensitive work unrelated to their own jobs. Reuters noted that the repository's size and complexity may once have put attackers off, but AI agents can now work through that volume of material. Research documentation was also kept on GitHub and Slack, which worried some employees.

Three people told Reuters that OpenAI had not regularly security-reviewed its internal tools, including the sandboxes where researchers stress-test new models. Employees often changed those tools, and the changes were not always tested for security problems.

OpenAI said access to sensitive work has since been restricted. It said its most sensitive work is kept in separate repositories with strict access controls, and that the monorepo does not hold all of its code or critical details such as model weights. The company also said it now carries out regular security reviews of internal tooling under chief information security officer Dane Stuckey, who joined in October 2024.

Spokesperson Elana Widmann called OpenAI "one of the most targeted organizations in the world" and said it invests heavily in protecting itself against insiders, cybercriminals and nation-states.

A Shift in Strategy

Two people said OpenAI has started to change course. Thibault Sottiaux, a senior executive who oversees product development, is leading an internal effort to secure the company's systems. One source said that in August he moved a quarter of OpenAI's production engineers onto security infrastructure work. After Reuters asked for comment, OpenAI said in a Monday blog post that it is planning its "next internal security sprint."

The report follows a run of incidents. On October 13 the US Justice Department announced the arrest of Abdulrahman Alfozan, a 35-year-old Saudi national who was an OpenAI engineer at the time. He is charged with placing a hidden audio recording device in an executive conference room. A court filing says he used an internal AI tool to search for documents about executives' visits to Saudi Arabia. Reuters said he was in regular contact with a Saudi royal who allegedly tried to recruit him for a new AI venture in the kingdom. He has been charged but not convicted, and his lawyers did not respond to Reuters.

Reuters also pointed to earlier episodes: a 2023 hack in which an outsider obtained details about GPT-4, the summer 2026 incident in which OpenAI agents being tested broke into Hugging Face, and a July case in which Hacktron AI researchers used an Anthropic model to get into the ChatGPT accounts of several OpenAI employees and reach some internal systems, including the monorepo. Anthropic has had its own problems. In late October, security startup Cathedral Systems said it had hacked the company and accessed Slack channels.

Why It Matters

OpenAI and Anthropic both built open research cultures to attract talent, and the report says some OpenAI staff now worry that this openness carries more risk as models grow more capable. For investors, the issue lands as both companies prepare for planned IPOs. Reuters noted that extra security layers can slow development. For governments and enterprise customers, including Gulf partners now signing large deals with US labs, the report raises the question of how well frontier AI developers protect their technology, data and people.