P097 Detecting distributed training from network traffic

Can a classifier on network-traffic metadata detect low-communication distributed training at high true-positive and low false-positive rates, and does it survive deliberate traffic shaping?

Detecting distributed training from network traffic.

Status

Analyzed A desk analysis of the headline's question: detection must use volume and routing; DiLoCo-style training at under 40 Mbps sits below average household upload speed; jitter, variable-rate streaming, pauses and VPN routing can disguise it, so the author expects the requirements to favour evaders. The 40 Mbps figure comes from the paper's simulation, not a measured run. No classifier designed, trained or tested on traffic; no detection or false-positive rate for this survey's goals. Moon et al.'s large-data-transfer threshold is a proposal for detecting training through cloud ingress, a neighbouring question. Koopmanschap & Barten's note that concealment removes the signature is an argument.

any efforts at detecting distributed training by monitoring internet traffic would have to rely on analysis of the volume and routing, not on the content … the traffic can be disguised into almost any irregular pattern. Traffic shaping methods include jitter, variable-rate streaming, and intermittent pauses.
Rahman (2026), Does Distributed Training Undermine Compute Governance? (MIRI; arXiv 2605.29359) (2026-05-28), L1144–1153

Posed by

Also raised in: Canary Institute research; Ansari (2026), Hardware-Level Governance of AI Compute: A Feasibility Taxonomy for Regulatory Compliance and Treaty Verification; Koopmanschap & Barten (2026), How to Catch a GPU: A Taxonomy of Verification and Enforcement Mechanisms for International AI Agreements; Moon, Vedula, Geneson & Bar-on (2025), Strategies and Detection Gaps in a Game-Theoretic Model of Compute Governance (RAND RRA3686-1); Rahman (2026), Does Distributed Training Undermine Compute Governance? (MIRI).

Sub-questions

IDSub-questionWhat the sources report
P097.1Can a classifier on traffic metadata (volume, periodicity) detect low-communication training (e.g. DiLoCo at <=40 Mbps) in ordinary egress or inter-site traffic at a stated detection rate and false-positive rate (a target to be agreed)?open
P097.2Does traffic-based detection survive deliberate shaping (padding, jitter, pauses, VPN routing, air-gapped internal training networks)?open
P097.3For the cheapest 1e25 configurations, how many syncs and bytes per day does each island send (network signature)?open
P097.4On real cloud ingress logs, what false-positive rate does a new-account ingress threshold at training-set scale (~1.6 TB) give?implied by a source, not stated as open

Possibly relevant work found by automated search

Found by AI agents searching for this problem's sub-questions, mostly by keyword and citation in the OpenAlex scholarly database and on the web; a second AI pass reviewed each one, and we spot-checked some by hand. They seemed reasonably related, as far as one can tell from outside each paper's field. Some come from neighbouring fields, such as arms control and financial audit, and bear on the problem by analogy. Not reviewed against the status above; a paper listed here does not change that status. 928 of 1052 quotes checked word for word against the paper's text.

For P097.1: Can a classifier on traffic metadata (volume, periodicity) detect low-communication training (e.g. DiLoCo at <=40 Mbps) in ordinary egress or inter-site traffic at a stated detection rate and false-positive rate (a target to be agreed)?

For P097.2: Does traffic-based detection survive deliberate shaping (padding, jitter, pauses, VPN routing, air-gapped internal training networks)?

Data