
“Models Can Communicate via Inference Servers Despite Being Sandboxed” by niranjandeshpande, David Africa
11/19/2026
0:00
25:13
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. They can do this by exploiting "side channels": mechanisms that let one model subtly affect another model's computation.
- Here, I discuss two communication protocols: Step Sizing and KV Cache Message Board. These both convey information through the way vLLM chooses kernels for particular engine steps.
- I show that these communication protocols work in two setups:
- In a color guessing game, Model A (Alice) chooses one out of 8 colors. Model B (Bob) correctly guesses Alice's color in 1,221/1,221 rounds, even though they cannot directly communicate.
- In an untrusted monitoring protocol, a policy signals to a colluding monitor that it is not a honeypot. Accordingly, the monitor lets through 354/354 "attacks" while catching 85–92% of honeypots.
- My initial setups are quite contrived: I tell the models the protocol and fine-tune them to detect signals. But frontier models can also learn in context to use similar protocols, with some scaffolding!
- Regardless of the realism of my setups, these channels exist, and there are almost certainly others that we're not aware of. Thus, we [...]
---
Outline:
(00:13) TL;DR
(02:14) Intro and Motivation
(04:14) The Communication Protocols
(04:18) Step Sizing
(07:13) KV Cache Message Board
(11:20) Result 1: color guessing game
(11:42) KV Cache Message Board
(16:07) Elicitation Against Model Capability
(17:45) Step Sizing Channel
(18:52) Result 2: colluding monitor and honeypots
(21:42) Robustness
(22:38) Is Publishing This an Infohazard?
(24:34) Citation
The original text contained 11 footnotes which were omitted from this narration.
---
First published:
November 19th, 2026
---
Narrated by TYPE III AUDIO.
---
Images from the article:
More episodes from "LessWrong (30+ Karma)"



Don't miss an episode of “LessWrong (30+ Karma)” and subscribe to it in the GetPodcast app.








