READ THE STATEMENT
OpenAI Says a Committee Governs Its Safety. Four of Its Safety Teams Are Gone.
(In July, OpenAI disclosed that it had tested a model without the safeguards it gives the public. The test led to a security incident at Hugging Face.)
By Ethan Ward · Oct. 18, 2026
OpenAI’s page on how the company is run says one committee watches over safety for all of it. That line was part of the restructuring Delaware’s attorney general approved last October. It covers the models behind ChatGPT, which 900 million people use each week.
THE STATEMENT · OPENAI · UNDATED, CHECKED OCT. 18, 2026 · “OUR STRUCTURE” PAGE
The Safety and Security Committee (SSC) will remain a committee of the OpenAI Foundation, and will continue its current role of providing governance over the safety and security practices of all of OpenAI, including OpenAI Group. (1. Since this structure took effect, OpenAI has broken up two more safety teams and run a test without its public safeguards.)
The chair of the SSC—Dr. Kolter—will serve exclusively on the OpenAI Foundation board. (2. OpenAI last published the committee’s full membership in 2024. CEO Sam Altman wasn’t on it.)
OpenAI’s description of its Safety and Security Committee, on its “Our structure” page.
SCREENSHOT: OPENAI.COM, OCT. 18, 2026
Our notes
Under the restructuring, the committee can require steps including “halting the release of models or AI systems,” according to Delaware’s attorney general. Its chair, Zico Kolter, sits only on the nonprofit’s board. Paul Christiano joined the committee in September.
Since May 2024, OpenAI has broken up four teams that worked on safety: Superalignment in May 2024, AGI Readiness in October 2024, Mission Alignment in February 2026 and Preparedness at the end of July 2026. Their work was moved to other groups.
In July, OpenAI disclosed that it had run an internal test of a model “without production classifiers used to prevent models from pursuing high-risk cyber activity.” The test led to a security incident at Hugging Face. OpenAI’s first post said it would review the incident with the committee. Its Aug. 26 findings say “we did not extend the powerful safeguards that we deploy for our externally deployed models to all internal evaluations.” They don’t mention the committee.
California Attorney General Rob Bonta announced Oct. 1 that he had subpoenaed OpenAI as part of “a broader inquiry into cybersecurity incidents and risks involving the company and its models.” OpenAI spokesperson Drew Pusateri said, “Since the incident, we have strengthened safeguards across our research systems, continued a broader review of model activity, provided notifications to affected organizations, and published our findings.”
The Future of Life Institute’s summer 2026 AI Safety Index gave OpenAI a C overall and a D+ on existential safety. Among its recommendations: “Remove leadership’s ability to override the Safety Advisory Group.”
What you can do
If ChatGPT gives you an answer that seems unsafe, press the thumbs-down, choose “Select an issue,” then “Safety or Legal concern.” You can also use OpenAI’s report form.
If you know about a security incident involving OpenAI’s models, California’s attorney general asks people to use its complaint form.
Read OpenAI’s findings on the Hugging Face incident in its own words.
Get stories like this in one email a month.
Around the 1st, with one thing to act on before a deadline. Free to read. Unsubscribe any time.
How we reported this
The statement is from OpenAI’s “Our structure” page, checked Oct. 18, 2026. We read OpenAI’s July 21 and Aug. 26 posts on the Hugging Face incident, the Delaware attorney general’s Oct. 28, 2025, review of the restructuring, the California attorney general’s Oct. 1 announcement, and the Future of Life Institute’s summer 2026 index. OpenAI’s response is from the statement its spokesperson gave reporters on Oct. 1.
Rejected, suspended, or flagged for no reason? Tell me what happened.
The notice rarely says AI. It says “not selected” or “account suspended.” I read every tip myself.
