Physical Verification of AI Compute
As governments seek to enforce limits on AI training, verification methods based on physical signals such as power, heat and network traffic will recur as the practical alternative to trusting operators or chip vendors, and will be written into export-control and treaty proposals.
forming · confidence 40 · Emerging (watchlist) · tracking since October 23, 2026 · updated October 23, 2026
Why the conviction moved
- Oct 23Strengthened +2
On simulated data-center power data, a Viterbi tracker caught at least 96% of undisguised training runs at a 1% false-alarm rate, and at least 87% of runs over a day when operators tried to hide them. That is quantified evidence that facility power meters can verify training without trusting the operator. The trend likely opened on this same paper, so the confirmation is kept small.
Source trail
Supporting · October 23, 2026
Researchers Show Facility Power Meters Can Detect Hidden AI Training Runs Without Trusting the Operator
On simulated data-center power data, a Viterbi tracker caught at least 96% of undisguised training runs at a 1% false-alarm rate, and at least 87% of runs over a day when operators tried to hide them. That is quantified evidence that facility power meters can verify training without trusting the operator. The trend likely opened on this same paper, so the confirmation is kept small.
arXiv cs.CR
Unlock full source trail, score history, and daily updates.
Unlock TrendsTownsquare
Argue the thesis in Townsquare.