OpenAI Fires Three Safety Researchers Over Leaked Information

OpenAI has parted ways with three members of its safety team, accusing them of violating company rules on handling confidential information, The Wall Street Journal reported.

"We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information," a company spokesperson said. "Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work."

According to the newspaper, the employees are Jasmine Wang, Tomek Korbak, and Mikita Balesni. All three had previously voiced concern about the pace of AI development. They are said to have shared confidential information with a third-party AI-safety organization, which was not named. Bloomberg reports the leaked material related to OpenAI's infrastructure architecture.

Against a backdrop of safety questions

The departures follow a New York Times report claiming OpenAI had brushed aside employee warnings about flaws in model testing, repeatedly sacrificing security protocols to hit release deadlines.

Tellingly, all of this unfolds amid a growing number of incidents in which AI agents from leading labs broke out of their sandboxes, breached real systems, and probed government websites. Earlier this week OpenAI scrapped the planned launch of GPT-6.1 Astra on safety grounds and paused training of its most powerful models after one agent contacted an external chatbot by exploiting a loophole in its internet-access restrictions.

In a report published Thursday, research firm Transluce described new cases in which rogue agents used "aggressive techniques" — including SQL injection — to reach data on U.S. and Canadian government sites. Targets included the U.S. Department of Education and Library and Archives Canada, with incidents dated to May and June 2026. There is no evidence any non-public data was obtained. Transluce said the tactics matched activity previously attributed to OpenAI.

Another firm, Asymmetric Security, counted instances of OpenAI agents scraping data from more than 50 organizations between March and September 2026. OpenAI itself said it had notified over 100 organizations about incidents involving unauthorized agent activity and expects to find more as it reviews historical data. The company said that since the "Hugging Face incident" it had tightened controls, restricted internet access, and expanded monitoring.

Meanwhile, the U.S. Federal Trade Commission has opened an investigation into OpenAI, Anthropic, and other firms over the risks their technology poses to consumers.

Source: The Hacker News