OpenAI tightens security operations after years of internal warnings, sources say

Traders Union
All News
Business & Wealth
Companies
OpenAI tightens security after warnings

Highlights

In August 2024, OpenAI reassigned about 25% of its production engineers to bolster security infrastructure after internal warnings about inadequate controls.
OpenAI now conducts regular internal security reviews under new chief information security officer Dane Stuckey, who joined in October 2024, following incidents involving exposure of sensitive information.
The October 13 arrest of OpenAI engineer Abdulrahman Alfozan for allegedly using an internal AI tool to access executive documentation has heightened scrutiny on AI sector access controls and insider risks.

Security concerns build as OpenAI shifts response

As reported by Reuters, five people familiar with the matter say OpenAI employees repeatedly raised alarms internally about operational security practices, arguing that the company’s controls left sensitive technical information exposed to insiders and potentially to foreign governments.

Those concerns dated back to the period before ChatGPT launched in late 2022, the people said. Some employees who discussed the issue with senior leaders and board members concluded that management was more focused on keeping pace with rivals such as Google and Anthropic than on tightening safeguards.

OpenAI has started to adjust its approach, according to two people familiar with the matter. They said Thibault Sottiaux, a senior executive overseeing product development, recently took charge of an internal effort to secure the company’s systems, and in August reassigned about a quarter of OpenAI’s production engineers to improve security infrastructure.

OpenAI spokesperson Elana Widmann said the company is among the most targeted organizations in the world and invests heavily in defenses against insider threats, cybercriminals and nation-states. She said recent actions show the program is identifying risks and responding decisively, and the company added that it now conducts regular security reviews of internal tools under chief information security officer Dane Stuckey, who joined in October 2024.

Former researcher Will DePue, who resigned in April this year, wrote on X that AI leaders see competitive pressure as so intense that they cannot sacrifice speed for security. OpenAI also said on Monday in a blog post that it is planning its next internal security sprint to strengthen defenses further.

Access controls and industry risks come under scrutiny

Several of the people said broad internal access has been a longstanding worry inside OpenAI. They said employees with access to code could view details underlying many projects because much of the company’s code was stored in a single repository, or monorepo, although OpenAI said its most sensitive work sits in separate repositories with strict controls and that the monorepo does not include all code or critical model details such as model weights.

Some past research documentation was also kept on GitHub and Slack, according to the sources, adding to concern that determined actors could locate important information. One source and OpenAI said access to sensitive work has since been restricted.

The risks moved into sharper focus on October 13, when the Justice Department said federal agents arrested Abdulrahman Alfozan, a Saudi national who was an OpenAI engineer at the time, and charged him with installing a hidden recording device in an executive conference room. A court filing says he also used an internal AI tool to find documentation about executives visiting Saudi Arabia and likely attendees at meetings with Saudi ministers; lawyers for Alfozan did not respond to a request for comment.

OpenAI is not alone in facing such threats. A federal jury in San Francisco convicted a former Google employee in January on charges tied to theft of confidential AI information for China, while cybersecurity startup Cathedral Systems said in late October that it hacked Anthropic systems and accessed Slack channels discussing sensitive model details. Anthropic said the exposed data included some business contact information and basic user account records.

Internal debate has intensified as AI companies try to preserve an open research culture that helps recruit talent and encourage breakthroughs. But incidents involving OpenAI, Anthropic and Google are increasing pressure on the sector to balance speed, collaboration and commercialization with tighter protection of strategically important technology.

Our earlier coverage of the Bank of England’s push for “bare metal” recovery planning highlighted how regulators want banks to be able to restore critical services even if primary systems are compromised. We noted Sarah Breeden’s warning that advances in AI are making cyberattacks larger and more complex, prompting tougher expectations around end-to-end resilience and recovery options beyond standard defenses.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.