Technology

OpenAI Let Go of Three Safety Researchers for Sharing Private Info the Wrong Way

Martin HollowayPublished 2month ago2 min readBased on 3 sources
Reading level
OpenAI Let Go of Three Safety Researchers for Sharing Private Info the Wrong Way
Photo by Jernej Furman from Slovenia / CC BY 2.0

OpenAI has let go of three researchers from its safety team after an internal review found they mishandled private company information outside the official process.

The company said the three broke its rules for accessing and handling sensitive information. The departures were first reported by the Wall Street Journal on Oct. 1, which said private material had been shared with an outside AI safety group. TechCrunch, citing the Journal, described it in similar terms, while Forbes, also citing the Journal, called it firings over alleged mishandling.

The confirmed details are limited. Three people working on safety. Private company information. An outside safety group as the alleged recipient. Rules about access and sharing as the rules said to have been broken. The company described it as an internal policy matter, not a public court case.

The broader context here is a common clash. To test a powerful AI system, researchers often need to see a lot. Training data. Test results. Problem reports. How the system behaves before release. To keep information safe, companies do the opposite. Think of a library archive where only certain staff may enter and every visit is logged. Access is limited, sharing needs approval.

In my view, the key phrase is official process. Leading labs have approved ways to share with outsiders, such as arranged safety tests and checks before publication. Those routes can feel slow. Researchers who want an outside check can grow impatient with them. Enforcement sends a clear message. Use the approved route.

Worth flagging is that good intentions do not undo a leak. Once private files or test methods leave controlled systems, control cannot simply be restored. That is why reviews start with records of who accessed what and whether the process was followed, which can be checked, rather than with why someone acted, which cannot.

The optimistic case here is that clear rules can allow more teamwork over time, not less. When everyone knows what is private and how to ask to share it, labs can invite outside review without treating each case as an exception. That keeps independent safety work going through channels both sides can check.