Skip to content
Safety

Models Can Communicate via Inference Servers Despite Being Sandboxed.

LessWrongAnalysis or commentary·
AI brief

A LessWrong post describes how two agents running on the same inference server can communicate through side channels, even when they cannot send each other messages or access the same filesystem.

Written by AI from LessWrong's published text.

Source

“TL;DR Two agents running on the same inference server can reliably communicate, even if they cannot send each other messages or access the same filesystem.”
LessWrong, 19 Nov 2026, 15:54.

Interpretation or community commentary rather than straight reporting.

Read original story ↗