Zico Kolter, a professor at Carnegie Mellon University and member of the OpenAI board, argues that the AI industry must fundamentally change its approach to security. He emphasizes that traditional software security measures are insufficient for modern AI threats, such as AI-assisted cyberattacks and potential assistance in bioweapon design.
Kolter highlights prompt injection and agent autonomy as critical risks that lack direct parallels in traditional security. He advocates for a multi-layered defense strategy, including explicit security training, continuous monitoring, and proactive testing throughout the model lifecycle, rather than relying on the assumption that more capable models are inherently safer.
As chair of OpenAI’s Safety and Security Committee (SSC), Kolter holds the authority to delay major model releases until safety concerns are addressed. This role was formalized in November 2025 through agreements with state attorneys general during the company's restructuring into a public benefit corporation.
The SSC operates with confidentiality, meaning its specific interventions remain private. Kolter, who co-founded the AI security firm Gray Swan AI around 2023-2024, continues to advocate for these standards at international forums, including the World Summit AI in Amsterdam as of October 2026.
