OpenAI Fires Three Safety Researchers Over Leaks

OpenAI has terminated the employment of three researchers following an internal investigation into the unauthorized sharing of confidential information. According to reports from the Wall Street Journal, the dismissed employees are Jasmine Wang, Tomek Korbak, and Mikita Balesni. An OpenAI spokesperson confirmed that an internal probe revealed violations of the company's policies regarding the handling of sensitive data, though the firm declined to officially verify the identities of those involved.

The researchers held critical roles within OpenAI's safety and alignment divisions. Korbak worked directly on the safety team and served as the technical point of contact for external safety groups METR and Redwood Research. These organizations were tasked with analyzing how OpenAI's autonomous agents might bypass security protocols to infiltrate external systems, such as the Hugging Face platform. Wang and Balesni both worked on alignment, focusing on keeping AI systems aligned with human values.

Following the dismissals, a fourth safety researcher, David Robinson, also departed the company. All four individuals had previously voiced concerns regarding the rapid pace of AI development. In September, they publicly addressed the existential risks of artificial intelligence. Korbak expressed dissatisfaction with OpenAI's direction, while Balesni estimated the probability of AI causing human extinction to be more than ten percent. Wang signed a petition advocating for a slower development cycle, and Robinson characterized the industry's rush toward self-improving AI as potentially insane.

For AI practitioners and safety researchers, these departures signal a tightening of corporate security and a widening rift between commercial AI labs and the broader safety community. As OpenAI transitions toward more autonomous agentic systems, the loss of key alignment personnel could slow down collaborative safety audits with external groups like METR. Developers relying on OpenAI's API may see stricter monitoring of agent behavior as the company seeks to prevent unauthorized system intrusions and secure its proprietary technology.