What it means that OpenAI fired three safety researchers accused of sharing confidential information — October 2, 2026
· CompaniesAutomation
Radar Flash Edition: OpenAI fires three safety researchers accused of passing information to an external organization. What to check in your company's internal reporting channel.
Flash edition. OpenAI has fired three members of its safety team who are accused of passing confidential information to an outside AI safety organization. The story broke after we closed this morning's Radar.
What happened
The Wall Street Journal broke the news on Thursday, October 1: three OpenAI safety researchers have left the company, allegedly for sharing confidential information with an outside organization (Source). OpenAI confirmed it in a single sentence: "Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work" (Source).
The company hasn't said what the information was or who received it, and the three haven't spoken publicly (Source). This comes at a bad time: days earlier, The New York Times reported that management had ignored internal safety warnings, and it's not the first time: in 2024, OpenAI already fired two safety researchers for an alleged leak (Source). Meanwhile, the FTC is preparing orders to compel OpenAI to testify, as we reported yesterday.
Why it matters
The line between "external evaluation" and "leak" isn't drawn by intent; it's drawn by procedure: what can be shown, to whom, and through which channel. OpenAI says its employees crossed it. The pattern is familiar: when someone who sees a problem believes their warning isn't reaching anyone internally, the problem goes outside and ends in dismissal, not a fix. It's the second blow to an AI lab's safety team in a month, after Jacob Coxon's resignation from Anthropic.
For your company
The risk starts when someone on your team sees an automation mishandling customer data and doesn't know who to tell. In Spain, for companies with 50 or more workers, Law 2/2023 already requires an internal reporting system, bans retaliation against anyone who uses it, and makes not having one a very serious offense, with fines of 600,001 to 1,000,000 euros. Article 87 of the EU AI Act points to the same directive that this law transposes for reporting breaches of the Act. Three checks:
- Make sure your channel covers AI. Check that it takes reports about automations and personal data, not just fraud or harassment.
- What your evaluator sees. If a consultant or auditor reviews your systems, put in the contract what they can see, where they store it, and who publishes their findings.
- Who responds and how quickly. Every AI-related warning gets an owner and a written deadline.
Only with that channel, and someone who answers, can a process move up to UNATTENDED on the AI First ladder.
Frequently Asked Questions
Were the three researchers whistleblowers?
Nobody knows yet. OpenAI says they broke its policies and hasn't said more. Cite it that way, not as retaliation or as a proven leak.
Does this change anything for a company using ChatGPT?
Not in today's service. It does add strain on the team that decides when a model is ready to ship; ask your provider for written notice before they switch your model.
What if nobody answers in the internal channel?
The law lets the employee go to Spain's Independent Whistleblower Protection Authority. Better to fix it in-house.