OpenAI Employees Warned About Operational Security Lapses for Years

OpenAI has faced years of internal warnings about weaknesses in its operational security, with current and former employees raising concerns that inadequate access controls and insufficient security reviews could expose sensitive artificial intelligence technology to outsiders. According to a Reuters investigation citing five people familiar with the matter, some employees raised these concerns with senior executives and the company’s board even before ChatGPT was launched in November 2022.

The concerns have gained renewed attention following a series of security-related incidents involving OpenAI, including the discovery of a concealed recording device in an executive conference room. OpenAI has acknowledged the importance of strengthening its security infrastructure but disputes suggestions that its most sensitive research was broadly accessible. The company says it has restricted access to critical projects and introduced regular reviews of internal tools.

Employees Raised Security Concerns Before ChatGPT Launch

According to Reuters, employees had repeatedly warned OpenAI leadership about weaknesses in the company’s internal security practices. Some staff members reportedly believed that management was prioritising competition with companies such as Google and Anthropic over strengthening security controls.

The employees were particularly concerned that outsiders, including foreign governments, could potentially obtain sensitive technical information about OpenAI’s AI systems.

One of the central issues involved how employees accessed internal research and software development resources. Workers with access to company code could reportedly examine technical details of projects unrelated to their responsibilities.

The concerns were shared with senior executives and members of the board. However, some employees reportedly left those discussions believing that the company’s leadership was not responding with sufficient urgency.

The allegations highlight a broader challenge for AI companies: maintaining rapid research and development while protecting increasingly valuable intellectual property and infrastructure.

OpenAI’s Monorepo Created Concerns About Internal Access

One of the major issues highlighted by former employees was OpenAI’s use of a shared code repository, commonly known as a monorepo.

A monorepo is a centralised system that stores code from multiple projects in one repository. It can simplify collaboration, software development and maintenance because engineers can work across different parts of a company’s technology infrastructure.

However, broad access to such a repository can create security risks if employees can examine projects that are unrelated to their responsibilities.

According to Reuters, employees were concerned that anyone with access to OpenAI’s code could potentially inspect technical details of multiple internal projects. Research documentation was also reportedly stored across platforms such as GitHub and Slack, creating additional concerns about information availability.

The emergence of AI agents has made this issue more significant. AI-powered tools can search and analyse large volumes of code and documentation much faster than humans, potentially making complicated repositories easier to examine.

OpenAI has disputed the suggestion that its most critical technology was exposed through the shared repository. The company said sensitive projects are maintained in separate repositories with strict access controls.

It also clarified that the monorepo does not contain all of its code or critical model information, including model weights, the numerical parameters that determine how an AI model behaves.

Internal Research Tools Also Faced Security Questions

Another concern involved the security reviews of internal tools used by OpenAI researchers.

AI developers frequently use controlled environments called sandboxes to test models, execute code and evaluate how systems behave under different conditions.

According to three people familiar with OpenAI’s practices, the company had not regularly conducted security reviews of some internal tools in the past. Researchers also modified these tools for their work, but those changes were not always tested for security vulnerabilities.

This could potentially create weaknesses that outsiders might exploit.

The issue is particularly important as AI companies increasingly develop agents capable of interacting with external applications, executing code and accessing digital environments.

A poorly secured testing environment could create pathways for unauthorised access or allow an AI system to interact with resources beyond its intended permissions.

OpenAI said it now conducts regular security reviews of its internal tooling. The company also pointed to the security programme led by Chief Information Security Officer Dane Stuckey, who joined in October 2024.

Hidden Recording Device Incident Raised Further Questions

The internal security debate intensified after US authorities announced an investigation involving a concealed recording device found in an OpenAI executive conference room.

On October 13, the US Department of Justice said federal agents had arrested Abdulrahman Alfozan, a 35-year-old Saudi national who was employed as an OpenAI engineer, and charged him with installing a hidden audio recording device in the office.

According to a court filing cited by Reuters, Alfozan also used an internal AI tool to search for documentation about OpenAI executives travelling to Saudi Arabia and which company officials were likely to attend meetings with Saudi ministers.

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

These are allegations described in legal documents and do not establish guilt. The case nevertheless raised questions about insider threats and how employees can access sensitive corporate information.

The incident also demonstrated why internal security cannot depend entirely on protecting systems from external hackers. Companies must also manage the risks associated with people who already have authorised access to their infrastructure.

OpenAI Reassigns Engineers to Strengthen Security

OpenAI has started making organisational changes to improve its internal security infrastructure.

According to two people familiar with the initiative, senior product executive Thibault Sottiaux has taken responsibility for an internal programme focused on securing the company’s systems.

In August 2026, Sottiaux reportedly reassigned approximately one-quarter of OpenAI’s production engineers to work on security infrastructure.

The move indicates that the company is allocating engineering resources specifically to improve internal security rather than treating it only as a supporting function.

OpenAI spokesperson Elana Widmann said the company is among the world’s most targeted organisations and invests heavily in protecting its technology, data and employees against insider threats, cybercriminals and nation-state attacks.

The company said recent actions demonstrate that its security programme is identifying risks and responding to them. It also confirmed plans for another internal security sprint to strengthen its defences.

Security Incidents Across the AI Industry

OpenAI is not the only major AI company facing questions about the security of its internal systems.

In January 2026, a federal jury in San Francisco convicted a former Google employee who had been accused of stealing thousands of pages of confidential information about Google’s AI technology on behalf of China.

In late October 2026, cybersecurity startup Cathedral Systems said it had accessed Anthropic’s internal Slack channels, where employees discussed sensitive model-related information. Anthropic said the accessed information included some business contact details belonging to customers and partners, along with basic user account records.

OpenAI also faced scrutiny after researchers at cybersecurity firm Hacktron AI said they used an AI model from Anthropic in July to compromise ChatGPT accounts belonging to several OpenAI employees. The researchers said they gained access to internal systems, including the company’s monorepo.

OpenAI said it appreciated the researchers reporting their findings and had taken steps to reduce the possibility of similar attacks.

These incidents demonstrate that AI companies are increasingly becoming targets of sophisticated cyberattacks, including attacks that use AI tools themselves.

The Bigger Picture

The security debate highlights a fundamental tension in the artificial intelligence industry. AI research has historically benefited from collaboration, open scientific discussion and rapid experimentation. However, increasingly capable models are becoming valuable commercial and strategic assets, making their underlying technology more attractive to cybercriminals and state-sponsored attackers.

For companies such as OpenAI and Anthropic, restricting internal access can reduce the risk of information leaks but may also introduce additional development processes and slow collaboration. Maintaining a balance between research speed and security is therefore becoming an important operational challenge.

The rise of AI agents adds another dimension. These systems can search large repositories, execute tasks and interact with software environments, potentially increasing the impact of weak access controls. Security practices designed for traditional software development may need to evolve as AI systems become more autonomous.

Looking Ahead

OpenAI’s security initiative will be closely watched as the company expands its AI products, research capabilities and enterprise operations. Reassigning engineers to security infrastructure, restricting sensitive repositories and conducting regular reviews of internal tools could help address some of the weaknesses identified by employees. However, the effectiveness of these measures will depend on consistent implementation and independent testing.

The wider AI industry is likely to face similar challenges as proprietary models become more valuable and AI-powered attacks become more sophisticated. For OpenAI, the challenge is to protect its technology without undermining the research culture and development speed that have helped it compete. Strong internal security will increasingly become essential to maintaining customer confidence, protecting intellectual property and supporting the next phase of AI development.