OpenAI updates safety and security practices after o1 release

OpenAI announced changes to its Safety and Security Committee (SSC) four days after releasing o1-preview, converting a body that had previously included company executives into an independent oversight committee of the board. The committee is chaired by Carnegie Mellon machine-learning professor Zico Kolter and includes Quora co-founder Adam D’Angelo, retired US Army General Paul Nakasone and former Sony general counsel Nicole Seligman — none of whom hold an operational role at OpenAI.

The announcement followed a 90-day review the committee had conducted of OpenAI’s safety and security processes, after which it made recommendations to the full board. Under the new arrangement, company leadership briefs the committee on safety evaluations ahead of major model releases, and the committee — along with the full board — has the authority to delay a launch until concerns are addressed. OpenAI also said it was evaluating an industry-wide Information Sharing and Analysis Center for AI, to share threat intelligence and cybersecurity information across the sector.

The restructuring came against a backdrop of departures from OpenAI’s safety functions during 2024, including the dissolution of its “superalignment” team, and public criticism that a committee staffed partly by executives evaluating their own company’s releases amounted to self-policing. Making the committee independent, and giving it explicit power to pause a launch, was a direct answer to that criticism, though the committee’s decisions are advisory to a board that OpenAI’s leadership still sits on and its practical effect on any specific release was not tested publicly in the following months.