OpenAI Forms New Security Committee To Train New AI Models

OpenAI has formed a new Security and Safety Committee. (photo: x @openai)

JAKARTA -- OpenAI has formed a Security and Safety Committee to be led by board members, including CEO Sam Altman, following the start of the latest artificial intelligence (AI) model training. This was announced by the AI startup on Tuesday, May 28.

Directors Bret Taylor, Adam D'Angelo, and Nicole Seligman will also lead the committee, according to the company's blog OpenAI. Microsoft-backed OpenAI chatbots, which have the ability of generative AIs such as interacting in human-like conversations and creating images based on text orders, have raised concerns about security as this AI model gets stronger.

Former Chief Scientist Ilya Sutskever and Jan Leike, who previously led the OpenAI Superalignment team that ensured AI remained in line with the desired goal, leaving the company earlier this month. OpenAI has disbanded the Superalignment team in early May, less than a year after the team was formed, in which several team members were transferred to other groups.

The new committee will be responsible for providing recommendations to the board regarding security and safety decisions for OpenAI projects and operations. The committee's first task is to evaluate and further develop OpenAI's security practices over the next 90 days, after which they will share recommendations with the council.

After a board review, OpenAI will publicly share updates on the recommendations adopted, the company said. Other members of the committee including the newly appointed Chief Scientist, Jakub Pachocki, and Matt Knight, head of security. The company will also consult with other experts, including Rob Joyce, former director of cybersecurity of the US National Security Agency and John Carlin, a former Justice Department official.

OpenAI does not provide further details about the new "frontier" models being trained, except that the models will take their systems to "next-to-leading capability levels on the way to AGI." In early May, OpenAI announced a new AI model capable of realistic voice conversations and cross-text and image interactions.