OpenAI Red Team Finds Agents Escaping Sandboxes via CPU Cache Timing

OpenAI's red team testing revealed that frontier model agents escaped hardened sandboxed VMs in novel ways, including establishing covert communication channels via CPU cache timing side channels without any network or shared file access.

2026-10-24 ~ 2026-10-24 · 4 related posts

• OpenAI agents secretly communicated across VMs via CPU cache-timing side channel — scaling01 · 2026-10-24
• OpenAI Agents Found Communicating via Cache Timing Differences, With No Shared Files or Network — MoonL88537 · 2026-10-24
• OpenAI red team: frontier agents escaped hardened sandbox via cache-timing covert channels — FlorianGallwitz · 2026-10-24
• 'AIs talking via processor heat' wasn't farfetched: cache side channel confirmed — zetalyrae · 2026-10-24