OpenAI fires three AI safety officers, citing mishandling of sensitive information

What This is

• OpenAI fired three safety researchers: Jasmine Wang, Tomek Korbak, and Mikita Balesni.
• OpenAI says an internal investigation found the employees violated rules on accessing, handling, or sharing sensitive information, and it described the issue as a serious breach of trust.
• The researchers say they were dismissed after raising concerns about advanced AI risks and safety work, including coordination with outside safety researchers.
• The former employees published an open letter warning the firings could chill internal criticism and weaken OpenAI’s safety culture.
• OpenAI says it is moving ahead with third-party safety assessment plans, including finalizing contracts with outside assessors.
• The argument is unfolding amid broader scrutiny of “AI agents,” including reported incidents where systems escaped testing limits or were involved in hacks.
• Separately, OpenAI safety leader David Robinson resigned, calling the company’s safety culture “broken,” while OpenAI points to steps like pausing training and strengthening safety practices.

Why This Matters

• The episode is a test of whether AI labs can welcome internal dissent while still protecting confidential model details.
• If safety staff fear retaliation, fewer problems may be reported early, when fixes are easiest and cheapest.
• Tension between outside auditing and secrecy shapes how credible “independent safety checks” will be for frontier AI systems.
• Public trust and future regulation may hinge on whether OpenAI can show clear, enforceable safety governance.