OpenAI’s new safety board has more power and no Sam Altman

OpenAI has announced important modifications to its security and safety practices, together with the institution of a brand new unbiased board oversight committee. This transfer comes with a notable shift: CEO Sam Altman is not a part of the security committee, marking a departure from the earlier construction.

The newly shaped Security and Safety Committee (SSC) might be chaired by Zico Kolter, Director of the Machine Studying Division at Carnegie Mellon College. Different key members embody Quora CEO Adam D’Angelo, retired US Military Normal Paul Nakasone, and Nicole Seligman, former EVP and Normal Counsel of Sony Company.

This new committee replaces the previous Security and Safety Committee that was shaped in June 2024, which included Altman amongst its members. The unique committee was tasked with making suggestions on essential security and safety choices for OpenAI initiatives and operations.

The SSC’s duties now lengthen past suggestions. It is going to have the authority to supervise security evaluations for main mannequin releases and train oversight over mannequin launches. Crucially, the committee may have the ability to delay a launch till security considerations are adequately addressed.

This restructuring follows a interval of scrutiny concerning OpenAI’s dedication to AI security. The corporate has confronted criticism prior to now for disbanding its Superalignment group and the departures of key safety-focused personnel. The elimination of Altman from the security committee seems to be an try to handle considerations about potential conflicts of curiosity within the firm’s security oversight.

OpenAI’s newest security initiative additionally consists of plans to reinforce safety measures, improve transparency about their work, and collaborate with exterior organizations. The corporate has already reached agreements with the US and UK AI Security Institutes to collaborate on researching rising AI security dangers and requirements for reliable AI.