Nov 3, 2026 · 32 min
OpenAI firings expose the limits of voluntary AI oversight
OpenAI whistleblowers on the fight over AI safety | Machine Gods
The researchers’ account raises questions about whether frontier AI companies can investigate themselves while protecting employees who flag safety risks.
- 1Former OpenAI researchers link their firings to disputes over safety work, information handling, and alleged leaks.
- 2The episode argues that voluntary external audits can fail when companies restrict access and auditors face legal and institutional pressures.
- 3Makeda and Tomek call for independent oversight, monitorability standards, regulation, and workplaces where safety concerns can be raised without retaliation.
Don't miss
Makeda and Tomek explain why they believe voluntary external investigations cannot provide durable oversight when labs control access to evidence.
The brief
Makeda Belezny and Tomek Korbach describe being fired from OpenAI after investigations into alleged leaks and communications with outside researchers, while arguing that their safety work shaped the conflict.
The researchers recount an investigation involving Hugging Face, Meter, and Redwood Research, where limits on data access made it harder for outside experts to assess what happened.
Their account centers on a trust problem: companies need external scrutiny, but voluntary audits can collapse when access, confidentiality, and legal exposure remain under the lab’s control.
The discussion turns to monitorability as AI agents become harder to interpret, including whether systems that weaken safety checks are evading oversight or simply behaving unpredictably.
Makeda and Tomek argue for independent auditors, stronger monitorability standards, regulation, and an internal culture where employees can raise safety concerns without retaliation.
Books & mentions
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.