OpenAI forms safety committee as next frontier model begins training

OpenAI just confirmed that training for its next frontier model is officially underway, and it is building a new guardrail at the very top of the company to watch over it.

The company announced that its board of directors has formed a dedicated Safety and Security Committee. The group is tasked with making recommendations to the full board regarding critical safety and security decisions across all OpenAI projects and operations.

A 90-day sprint with national security muscle

The committee is led by board members Bret Taylor, Adam D’Angelo, Nicole Seligman, and CEO Sam Altman. Following his appointment to the board, retired U.S. Army General Paul M. Nakasone also joined the committee.

Several key technical and policy leaders are on the roster as well, including Aleksander Madry, Lilian Weng, John Schulman, Matt Knight, and Chief Scientist Jakub Pachocki. OpenAI will also consult outside cybersecurity figures, including former cybersecurity officials Rob Joyce and John Carlin.

The committee’s first assignment is an immediate 90-day evaluation of the company’s existing safeguards and processes. Once the 90 days are up, the committee will present recommendations to the full board, and OpenAI plans to release an update on which measures it adopts.

Why it matters

First, this confirms what everyone has been waiting to hear: OpenAI is actively training its next-generation frontier model to push toward AGI. For developers and businesses building on top of OpenAI, that means the clock has started on the next major capability leap.

Strategically, bringing in high-profile cyber defense officials and a retired four-star general like Nakasone is a clear effort to shore up credibility in Washington. As AI models inch closer to autonomous cyber offensive capabilities, national security readiness is becoming the primary battleground for regulation.

However, we think the governance structure itself raises valid questions. Sam Altman sits on the very committee meant to police the risks of the models his executive team is racing to ship. When push comes to shove between product launch deadlines and safety red flags, having the CEO review his own safety posture leaves an obvious conflict of interest that 90 days of internal reviews might not entirely solve.