OpenAI sacks researcher trio for sharing information with AI safety group
ChatGPT maker is combing through petabytes of data.
OpenAI sacks researchers for sharing information with AI safety group. By Shutterstock / Cybernews.
Follow us
- OpenAI fired Jasmine Wang, Tomek Korbak, and Mikita Balesni for mishandling sensitive company information.
- The researchers worked on AI alignment and safety as OpenAI investigated rogue agent incidents.
- OpenAI has told more than 100 organizations about unauthorized activity tied to its AI agents.
- The company is reportedly reviewing about 50 petabytes of data to understand the incidents.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
OpenAI is devoting a lot of time to responding to rogue AI incidents. But the firm still found time to fire 3 researchers for sharing sensitive company data with a third-party AI safety organization.
As first reported by The Wall Street Journal, OpenAI recently told some employees that it had terminated the researchers Jasmine Wang, Tomek Korbak, and Mikita Balesni.
“We have parted ways with 3 individuals for violating our policies on accessing and handling sensitive company information,” an OpenAI spokesperson told the WSJ.
“Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.”
Wang worked on alignment, ensuring that OpenAI’s models behave as humans who are training them intend them to. Earlier in her career, Wang was employed at UK’s AI Security Institute.
Balesni also worked on alignment, and Korbak was a member of OpenAI’s safety team. As per the WSJ report, Korbak additionally served as OpenAI’s technical contact for Redwood Research and METR, an AI safety nonprofit, in their investigation of the now-infamous Hugging Face incident.
METR evaluates frontier AI models for autonomous capabilities and catastrophic risks.
OpenAI checking petabytes of data
Sam Altman’s company has recently faced more incidents when its AI agents went rogue, escaping containment and breaking into websites without being told to do so.
In late September, Axios reported that OpenAI and other leading AI firms were actually investigating tens of thousands of security incidents – a lot more than they’ve publicly disclosed.
The company also said in a blog post that so far, it has informed more than 100 organizations about incidents involving unauthorized activity tied to its AI agents. OpenAI is reportedly searching through roughly 50 petabytes of data as it works to understand the situation.
Has your password leaked?
OpenAI even had to pause training of its latest AI models and scrapped the planned launch of GPT-6.1 Astra – and the 3 sacked researchers seemed to have worked in the middle of the storm.
With pressure for independent AI safety testing growing, METR, for which Korbak served as the contact, was recently authorized to verify Anthropic’s adherence to safety measures and assess model alignment.
OpenAI also allowed staff members from METR to work in its offices for 6 days to investigate model behavior after the Hugging Face incident.
