OpenAI has severed ties with three researchers who worked on the company's safety teams, The Wall Street Journal reported, marking the latest round of departures from the artificial intelligence lab's safety operations.
The exits add to a growing list of safety-focused staff who have left OpenAI over the past two years, a period in which the company has shifted toward shipping commercial products faster while facing scrutiny over how it manages the risks of its models.
OpenAI has not publicly commented on the individual departures. The company has previously said its safety work remains central to its mission and points to its Preparedness Framework, a set of protocols adopted in late 2023 for evaluating and mitigating risks from its most capable models before release.
Those protocols came under internal strain in May 2024, when OpenAI dissolved its long-term safety team, known as Superalignment, following the departure of co-founder Ilya Sutskever and researcher Jan Leike. Leike said at the time that safety culture at the company had taken a back seat to product development.
Several other safety leaders have left since. Co-founder John Schulman departed for rival Anthropic in August 2024, and policy researcher Miles Brundage announced his exit the same month, writing that he wanted to work with less restriction on what he could publish.
The turnover matters beyond OpenAI's walls. Its GPT models underpin products used by hundreds of millions of people each week, and rivals including Google, Anthropic, and Meta often follow its lead on how aggressively to test systems before release. Fewer experienced safety staff can slow that testing, according to outside researchers who track the industry.
OpenAI is simultaneously restructuring. The company is converting its for-profit arm into a public benefit corporation while its nonprofit parent retains control, a change that drew lawsuits from co-founder Elon Musk and objections from some former employees.
The company has also been hiring. OpenAI said in 2024 it had grown to roughly 3,000 employees, up from about 375 in 2022, and it continues to recruit for safety and policy roles as it builds out its frontier research teams.
