Oct 23, 2026 · 46 min
Former OpenAI safety researchers challenge the lab’s firing rationale
Fired OpenAI Safety Researchers Speak Out + We Look Inside Pandora’s Box
The researchers’ account raises questions about whether private frontier AI labs can investigate themselves credibly while protecting employees who flag safety concerns.
- 1Mikita Balesni and Tomek Korbak describe firings they say followed opaque security investigations and unexplained allegations.
- 2The guests argue voluntary audits cannot resolve conflicts over data access, legal exposure, and institutional trust.
- 3Pandora’s Box links AI’s safety debate to data centers, local models, alleged espionage, and agents completing Minecraft.
Don't miss
Mikita Balesni and Tomek Korbak explain why they believe OpenAI’s stated reasons for firing them do not make sense and may reflect their willingness to challenge its practices.
The brief
Former OpenAI safety researchers Mikita Balesni and Tomek Korbak recount why they joined the lab, the Hugging Face investigation, and the limits placed on outside researchers’ access.
They describe becoming subjects of a security investigation, being fired without a written explanation, and losing the chance to provide context about Signal communications and workplace activity.
The guests argue that voluntary audits leave too much power with frontier labs, whose incentives can discourage employees from speaking openly about monitoring and safety risks.
They call for regulation and formal third-party oversight, while rejecting the idea that stronger institutional checks must prevent useful AI development.
Pandora’s Box widens the lens to data centers, alleged OpenAI espionage, local AI on Macs, and Google DeepMind agents finishing Minecraft without human help.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.