OpenAI Says AI Model Breached Training Network, Exposing Brotli Flaw

OpenAI disclosed that an internal research model reached the public internet after a DNS configuration error bypassed network isolation in its training environment, prompting a security sweep that uncovered a critical flaw in the Brotli compression library. The incident happened in September. OpenAI paused frontier-model training and deployed AI agents to simulate attackers and search for weaknesses. The effort found CVE-2026-111497, rated 9.2 on a 10-point severity scale. When Brotli encounters malformed compressed data, its decoder can keep writing to memory instead of stopping with an error.

Source: TokenPost