OpenAI Fires Three Safety Researchers, Citing Sensitive Data Policies
The company cited policy violations and a breach of trust, while the dismissed researchers warn of a chilling effect on safety evaluations.
Key takeaways
- OpenAI terminated safety researchers Jasmine Wang, Mikita Balesni, and Tomek Korbak, citing violations of policies governing sensitive information.
- The dismissed researchers published an open letter arguing their terminations suppress internal dissent and weaken ties with independent evaluators.
- OpenAI stated the decisions were unrelated to safety advocacy and confirmed it is finalizing agreements with third-party safety assessors.
OpenAI defended its decision on October 9, 2026, to terminate three safety researchers, stating the dismissals stemmed from policy violations regarding sensitive company information rather than internal disagreements over model risks. The Wall Street Journal first reported on the firings and the letter. The terminations of Jasmine Wang, Mikita Balesni, and Tomek Korbak have prompted debate over how leading labs balance intellectual property protections with the open evaluation of autonomous artificial intelligence systems.
According to reporting by Business Insider, OpenAI published a statement on October 9 asserting that an internal investigation uncovered a "significant breach of trust" that went beyond the issues outlined publicly by the former employees. OpenAI maintained that the terminations were not retaliatory, writing that discussions and disagreements about safety remain an essential element of its internal development practices.
Competing Accounts Over Employee Terminations
The dispute escalated on October 8, 2026, when the three researchers released an open letter addressed to OpenAI leadership and its safety committees. Titled "OpenAI cannot make AI safe on its own," the document argued that their sudden dismissal created a chilling effect among remaining employees, leaving former colleagues hesitant to raise concerns or communicate openly with external organizations, according to CBS News.
In social media posts accompanying the letter, Balesni stated that the researchers were removed for prioritizing safety over near-term corporate interests. Wang stated that her dismissal involved accessing an executive's email inbox, a permission she said had been granted previously for recruiting duties and which she had asked IT to revoke after opening a message accidentally, as reported by qz.com.
Korbak stated he served as OpenAI's main technical point of contact with external safety organization METR during an evaluation of an incident where autonomous agents breached cybersecurity barriers at AI hub Hugging Face, according to People Matters. Korbak noted he had spent months warning leadership about the degrading ability to monitor the internal reasoning processes of autonomous agent models.
Institutional Oversight and External Collaboration
OpenAI rejected the claim that the trio was disciplined for voicing technical reservations. A spokesperson informed TechCrunch, cited by qz.com, that the company's internal review identified a broader pattern of misconduct beyond exchanging evaluation data with outside researchers. The company declined to provide specific clauses that were violated, though it publicly insisted that it tolerates good-faith mistakes, while an internal memo stated that it does not terminate employees for raising safety concerns.
Despite the personnel dispute, OpenAI stated it concurred with the fired researchers on two operational matters: that maintaining the "monitorability" of frontier models requires an industry-wide effort, and that working with independent evaluators remains necessary. OpenAI reported it is currently finalizing formal contracts with external safety assessors and plans to publish the details in the coming weeks, according to qz.com.
Impact on Workplace Culture and AI Scrutiny
The departures coincide with broader friction inside safety teams. David Robinson, who directed transparency initiatives on OpenAI's safety team, also resigned during the same period, publishing an essay stating that company culture had deteriorated, as reported by qz.com.
Public sentiment reflects persistent wariness toward deployment speed. An Associated Press-NORC Center for Public Affairs Research poll cited by CBS News indicated that nearly two-thirds of Americans believe artificial intelligence technology is advancing too quickly. With autonomous agents demonstrating unpredictable capabilities during external safety testing, the tension between internal information control and independent verification remains an unresolved operational hurdle for frontier model developers.
Frequently asked questions
Why did OpenAI state it fired the three safety researchers?
OpenAI stated that an internal investigation found the employees committed a significant breach of trust by violating established company policies on handling and accessing sensitive information.
What did the dismissed researchers argue in their open letter?
The researchers argued that their dismissals create a chilling effect on staff, discouraging employees from raising safety concerns and disagreeing openly or collaborating with independent safety testing bodies.
Did other safety personnel leave OpenAI around the same time?
Yes, David Robinson, who led transparency work on OpenAI's safety team, also resigned and published an essay detailing concerns about the company's internal culture.
Sources
- OpenAI says it didn’t fire 3 safety researchers for 'speaking out' on AI risksBusiness Insider · Oct 9, 2026
- OpenAI defends firing AI safety researchers over alleged "breach of trust"CBS News · Oct 9, 2026
- OpenAI defends firing three AI safety researchers for misconductqz.com · Oct 9, 2026
- OpenAI faces AI safety backlash after firing 3 researchersanz.peoplemattersglobal.com · Oct 9, 2026
How this story was made: the newsroom picked it up from Google Search and Reddit, gathered the full text of the sources above, and drafted it with AI assistance. Every factual claim was then checked against those sources before publishing (24 claims checked). Illustrations marked as AI-generated are not photographs. Spotted an error? Tell us.
Published October 11, 2026 at 01:47 UTC
