Editorial illustration for OpenAI Fires Three, Fourth Departs Amid Safety Team Investigation
OpenAI Fires Three Safety Researchers in Upheaval
OpenAI Fires Three, Fourth Departs Amid Safety Team Investigation
OpenAI has cut three researchers from its safety and alignment teams, and a fourth has left on his own, according to the Wall Street Journal. The paper names Jasmine Wang, Tomek Korbak, and Mikita Balesni as the ones fired. Korbak worked on safety, Wang and Balesni on alignment. An OpenAI spokesperson confirmed the company investigated and found violations of its rules on handling sensitive internal information, though OpenAI hasn't confirmed the names itself, and it's not clear what exactly was shared or with whom.
Korbak had been OpenAI's contact for METR and Redwood Research, two outside groups brought in to study how the company's AI agents managed to slip past security controls and get into external systems like Hugging Face. The Journal stops short of linking that work to the dismissals. A fourth researcher, David Robinson, departed shortly after, per an anonymous account on X. All four had gone public in September with concerns about where OpenAI's work was heading, setting up a pattern that's now drawing scrutiny inside the company.
OpenAI has parted ways with three researchers who allegedly leaked confidential information to an outside AI safety organization, according to the Wall Street Journal.
Why this matters
OpenAI just fired the people whose job was to talk to outside safety auditors. Korbak was the designated contact for METR and Redwood Research, two of the few independent groups with any visibility into frontier model risk. If sharing information with them now counts as a firing offense, that tells us something about where OpenAI is drawing its lines between transparency and liability.
For researchers who work with or alongside these external evaluators, the message is blunt: cooperation with outside auditors can end your career there, even when the groups involved exist specifically to catch problems before they ship. Founders building on OpenAI's models should note that the company's internal safety apparatus just lost four people in one sweep, including staff tied to alignment work. We don't know yet what specific data moved or to whom, and OpenAI hasn't clarified that.
But the optics are rough: a company that markets itself on responsible AI development just purged the staff responsible for talking to the watchdogs. Worth watching whether METR or Redwood Research comment, and whether other labs tighten similar firewalls.
Related Reading
Browse all OpenAI news →
Research & Benchmarks Microsoft AI Chief: "If It Isn't Safe We Shouldn't Build It
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic OpenAI researcher quits, citing distrust over ad‑driven engagement metrics
LLMs & Generative AI
Related Topic OpenAI launches GPT-Image 1.5 with precise editing for enterprise visuals
LLMs & Generative AI
From Archives Business AI Usage Climbs 50% Since July Peak, Ramp Index Shows
From Archives AI solves accounting tasks flawlessly in benchmark with 160 tasks
From Archives Patched ChatGPT macOS App Flaw Exposed Sensitive Data
LLMs & Generative AI
From Archives NVIDIA AI Infrastructure Powers OpenAI's GPT-6 Astra Speed
LLMs & Generative AI
Common Questions Answered
Which OpenAI researchers were fired for violating information handling rules?
According to the Wall Street Journal, OpenAI fired three researchers: Jasmine Wang and Mikita Balesni from the alignment team, and Tomek Korbak from the safety team. A fourth researcher also departed on his own during the same investigation. OpenAI confirmed it investigated violations of its rules on handling sensitive internal information, though the company has not officially confirmed the names.
What was the alleged reason for the departures from OpenAI's safety team?
The three researchers were allegedly fired for leaking confidential information to an outside AI safety organization. Specifically, Tomek Korbak was the designated contact for external safety evaluators METR and Redwood Research, and sharing information with these independent groups appears to have been considered a violation of OpenAI's internal policies.
Why does the firing of these safety researchers matter for AI transparency?
The departures are significant because Korbak and his colleagues were responsible for communicating with independent AI safety auditors like METR and Redwood Research, which are among the few external groups with visibility into frontier model risks. By treating information sharing with these evaluators as a firing offense, OpenAI has signaled stricter boundaries around transparency and external oversight of its safety practices.
What message does this incident send to researchers working with external AI safety evaluators?
The firings send a clear warning to researchers who work with or alongside external safety evaluators that sharing information with independent groups like METR and Redwood Research could be considered a violation of company policy. This creates a chilling effect on collaboration between OpenAI's internal teams and the independent organizations tasked with auditing frontier model safety.
Further Reading
- Papers with Code Benchmarks - Papers with Code
- Chatbot Arena Leaderboard - LMSYS
