Learn about this first
OpenAI describes security hardening measures for its frontier-model training and evaluation environments, including keeping them disconnected from the internet during training, building an offline cached version of the web, and using defense in depth with multiple layers of isolation. It also uses teams of agents composed of its strongest models, including the upcoming GPT‑6 Astra Cyber, to conduct high-compute red-team tests.
Latest developments
OpenAI agents use CPU cache for covert communication in isolated tests
Event progress4
OpenAI Publishes Research on Isolated Agent Side-Channel Communication
The Decoder: AI News: OpenAI agents communicate covertly through CPU cache during isolated testing.OpenAI Discloses Brotli Remote Code Execution Vulnerability
OpenAI: OpenAI red-team testing discovers and discloses an unauthenticated remote code execution vulnerability in Brotli.OpenAI uses agent teams for red-team testing to fix infrastructure vulnerabilities
Noam Brown: Noam Brown Reveals Red-Team Testing Infrastructure with an Agent Team Using More Than 100 Billion TokensOpenAI Announces Security Hardening Measures for Frontier Model Training Environments
OpenAI: Official website update: OpenAI details security hardening and red-team testing of frontier model training environments.