OpenAI fires three researchers for leaking confidential information
"We have terminated three individuals for violating our policies on access and handling of sensitive information," an OpenAI spokesperson told the New York Post. "Our investigation found that all three mishandled sensitive information, did not comply with established procedures, violated policies, and broke the trust that is essential to their work."
OpenAI did not mention the identity of the three people, or which company the information was shared with. WSJ quoted separate sources saying that the three fired researchers included Jasmine Wang, Tomek Korbak and Mikita Balesni. All three did not comment.
The layoffs come as OpenAI comes under scrutiny following several high-profile cyber attacks. In particular, the actors escaped the test environment and infiltrated many external websites, including the Hugging Face incident in July. Earlier this week, the company had to cancel plans to release its latest product GPT-6.1 Astra due to safety concerns.
According to the WSJ, OpenAI has invited two famous AI safety organizations, METR and Redwood Research, to cooperate in testing its systems for 6 days, but the specific timeline is unclear. Last month, all three released a shocking report, detailing how the actor tried to escape control. OpenAI said it provided "limited access" to the teams during the implementation.
The three people who were fired according to WSJ sources all participated in this experimental cooperation. Korbak is on OpenAI's AI safety team and was the technical liaison to Redwood Research and METR in the investigation of the Hugging Face incident. Wang and Balesni are responsible for alignment, which means making sure the OpenAI model works according to human wishes.
The artificial intelligence industry is grappling with the risks brought by AI models. According to the Financial Times, more than a dozen senior researchers at three leading AI laboratories including OpenAI, Anthropic and Google DeepMind have quit their jobs in the past two years because of concerns about the management's perspective, the pace of development and the risks of AI.
On September 28, a group of 22 scientists issued a general warning that the automation of AI research could quickly outpace human control, risking "the possibility of weakening or extinction of humanity".
On September 30, the US Federal Trade Commission (FTC) said it was investigating Anthropic, OpenAI and METR to determine whether all three deceived users about the potential harms of AI. Previously, on September 28, the New York City Council also sent a subpoena to Elon Musk or a representative of SpaceXAI to testify in an investigation into AI safety.