OpenAI is organizing a new internal security sprint that will bring outside researchers and security partners together with its own teams to use its models to test and harden systems. The company announced a partnership with Cathedral: Cathedral researchers will work alongside OpenAI teams, using OpenAI’s models to probe defenses, apply different expertise and approaches, and help identify weaknesses.
What the sprint involves
- Cathedral researchers will use OpenAI models in collaboration with OpenAI teams to explore potential vulnerabilities with varied techniques and perspectives.
- The initiative aims to find and fix weaknesses before they can be exploited, strengthening OpenAI’s defenses.
Changes to the Security Bug Bounty program
OpenAI is also updating its Security Bug Bounty program, more than quadrupling rewards for critical findings. The company is encouraging researchers to challenge assumptions and contribute to improving both systems and defensive methods.
Background and methods
This work is part of OpenAI’s ongoing effort to use AI to continuously discover and remediate weaknesses. In the previous sprint, the company allocated 25% of its product engineers to finding and fixing system weaknesses. Engineers used Codex to investigate reports, reproduce vulnerabilities in isolated environments, and prepare patches for review.
Why this matters now
OpenAI argues that traditional cyber defenses alone are no longer sufficient: recent industry incidents show how quickly threats can appear. The company emphasizes the need for continuous defensive loops to detect and fix weaknesses before attackers do.
Tools and support being offered
OpenAI says it is using its most capable models to harden its own defenses and is making those capabilities available to Daybreak customers. The company has also committed $1 billion in subsidized access over six months for frontline defenders that protect essential services and digital infrastructure.
OpenAI will share the architecture and lessons from the last sprint, along with a blueprint intended to empower cyber defenders across industry and government.
Next steps
OpenAI plans to continue advancing its models, applying them to strengthen its systems, and expanding access for trusted defenders. The company stresses that keeping the defender’s window open requires coordinated action across the industry and that work must begin now.
