OpenAI employees repeatedly warned senior leaders and the board that weak internal security could expose sensitive technology, with some concluding that keeping up with Google and Anthropic mattered more to management than fixing the problems, Reuters reported on November 3.
The concerns began before ChatGPT launched in late 2022, according to five people familiar with the matter. They covered broad access to company code, internal tools that weren't consistently checked for vulnerabilities and the risk that outsiders, including foreign governments, could obtain valuable information.
Employees questioned who could see sensitive work
The report describes a company trying to preserve a collaborative research culture while protecting technology that has become strategically valuable. Some employees believed its safeguards hadn't kept pace with that responsibility.
Will DePue, a former researcher who left in April, criticized the trade-off in an X post cited by the news agency:
"It is more important to them to move fast than to be secure, even from our greatest national adversaries."
One concern was OpenAI's shared code repository, or monorepo. People with access could see technical details from projects beyond their own work, the sources said. Its complexity may once have made it harder to navigate, but increasingly capable AI agents can search through large amounts of material.
OpenAI disputed the implication that its most valuable work was broadly exposed. It said sensitive projects are kept in separate repositories with strict access controls. The monorepo doesn't contain all its code or model weights, the numerical values that shape how an AI model behaves.
That distinction matters to the allegation: access to a shared codebase doesn't establish access to every model or its underlying weights.
Internal testing tools were another concern
Three people familiar with the company's practices said internal tools, including the sandboxes researchers use to test models, had not undergone regular security reviews in the past. Researchers often changed those tools for their work, but modifications weren't always checked for security weaknesses.
The company said it now conducts regular reviews of internal tooling under its security program, led by chief information security officer Dane Stuckey, who joined in October 2024.
The warnings also reached the board through a 2023 memo by then-researcher Leopold Aschenbrenner, according to three people familiar with it. He raised concerns about Chinese infiltration and later shared the memo with some directors after an outside hacker accessed company information. OpenAI said that intrusion was limited to Slack channels available across the company.
More recently, cybersecurity firm Hacktron AI said its researchers used an Anthropic model in July to access ChatGPT accounts belonging to several OpenAI employees and reach internal systems, including the monorepo. OpenAI said it thanked the researchers for reporting their findings and took steps to reduce the risk of similar attacks.
OpenAI points to a security push already underway
Two people said product-development executive Thibault Sottiaux had taken charge of an initiative to secure internal systems. One said he reassigned a quarter of OpenAI's production engineers to security infrastructure in August.
A new company post gives a separate account of that effort. OpenAI said its previous security sprint dedicated 25% of product engineers to finding and fixing weaknesses. Engineers used Codex to investigate findings, reproduce vulnerabilities in isolated environments and prepare patches for review.
The next sprint will bring outside researchers and security partners into that work. The company named Cathedral as a partner whose researchers will use OpenAI models alongside its teams to test defenses. It also said it was more than quadrupling bug-bounty rewards for critical findings.
Spokesperson Elana Widmann said the company invests heavily 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."
She argued that recent actions showed the security program responding to risks:
"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."