OpenAI Plans AI Security Sprint With Cathedral Systems

OpenAI is preparing an internal security sprint with Cathedral Systems and other researchers to use artificial intelligence models to test the company’s defenses and identify potential weaknesses, as it expands its bug-bounty rewards for security researchers. OpenAI said the planned effort will bring outside researchers together with its teams to challenge existing protections and test fixes. The announcement, published November 2, did not specify when the sprint will begin, which systems it will cover or what vulnerabilities it might uncover. No results from the planned work have been reported. The initiative builds on an earlier security sprint in which OpenAI said it assigned 25% of its product engineers to find and fix weaknesses. Engineers used the company’s Codex tool to investigate findings, reproduce vulnerabilities in isolated environments and prepare patches for review. OpenAI has described model-assisted security testing as an ongoing defensive process, rather than a one-time exercise. The company also said the earlier work exposed a deployment issue: a patch that had been merged did not always reach every deployed system. OpenAI said follow-up checks were needed,...