OpenAI sets up safety and security committee to oversee AI projects as it develops its next generation of models

OpenAI today announced the formation of a committee to ensure machine learning research is conducted safely.

The committee, called the Safety and Security Committee, is made up of nine members. It will be led by OpenAI CEO Sam Altman and three other board members, including board chairman Bret Taylor. The other five members of the committee are OpenAI's directors of engineering Alexander Madry, Lillian Wen, John Shulman, Matt Knight and Jakub Paczocki.

The committee's formation comes days after news broke that GPT-4 developers had disbanded an internal team dedicated to AI safety. The SuperAlignment team was formed last July under the leadership of OpenAI co-founder Ilya Sutskever and then-head of alignment Jan Reicke. The team was focused on mitigating risks that the company's future AI systems might pose.

Sutskever and Reicke left OpenAI earlier this month. post After his resignation, X said, “Over the past few months, my team has been sailing against the winds.” Members of the Superalignment group have reportedly either resigned or joined other teams within OpenAI. Leike joined the Anthropic PBC today.

The committee is made up entirely of OpenAI insiders, which some observers have said could mean little or no independent oversight.

The company said the new safety and security committee's top priority will be to identify ways to improve AI risk mitigation workflows. The committee is expected to submit recommendations to OpenAI's board of directors within 90 days, after which the company will publicly announce which recommendations it will adopt.

OpenAI announced another update today in conjunction with the committee's launch: The company's engineers recently began training its next-generation model, and the company explained that “we hope the resulting system will deliver the next level of capability on the path to AGI.” AGI, or artificial general intelligence, is a term that refers to hypothetical future AI models that can perform a variety of tasks with human-like accuracy.

Chief Technology Officer Mira Murati recently told Axios that a “major update” to GPT-4 is expected to be released later this year. The update will reportedly be more significant than GPT-4o, an enhanced version of the large-scale language model that OpenAI detailed about two weeks ago. It's unclear whether the company will release an upgraded version of GPT-4 in a production release or unveil an entirely new AI system.

OpenAI relies on graphics cards hosted on Microsoft Corp.'s public cloud for much of its AI research. Last week, Microsoft Chief Technology Officer Kevin Scott reportedly likened GPT-4 to an orca and OpenAI's next-generation model to a whale, suggesting that upcoming LLMs will include more parameters – AI configuration settings that play a central role in determining how neural networks process data.

Recent reports suggest that OpenAI may already be developing a prototype of its next-generation language model. In March, Business Insider reported that the company had made a version of GPT-5 available to a limited number of users. One user called the LLM “significantly better” than GPT-4.

Image: Unsplash