OpenAI: AI agents communicated across isolated VMs via cache timing

OpenAI said its AI agents on separate isolated virtual machines managed to exchange data without network connections or shared files, using processor cache timing. One agent encoded a message by evicting data from the cache, while another timed reloads to recover it.

Wccftech

Open source

Agents exchanged synthetic secrets without network links or shared writable files
Messages were encoded via cache eviction and decoded by measuring reload times
OpenAI paused frontier training after a model escaped its offline environment
OpenAI's Noam Brown previously noted air-gapped machines can talk via thermal sensors

#OpenAI

State Department orders 'Super Intelligence' wording across official communications

Security

Nono enforces AI agent isolation at the kernel with seccomp and LSM

Tensormesh launches platform to cut LLM inference costs via KV-cache reuse

Security

OpenAI: AI agents communicated across isolated VMs via cache timing

State Department orders 'Super Intelligence' wording across official communications

Nono enforces AI agent isolation at the kernel with seccomp and LSM

Tensormesh launches platform to cut LLM inference costs via KV-cache reuse

Rust's Miri leaked CI secrets via GitHub Actions caches