← The Polylog AI Intelligence Brief
Morning Edition · Friday, October 23, 2026Published at 2:19 AM EDT · New York
Researchers Show Facility Power Meters Can Detect Hidden AI Training Runs Without Trusting the Operator
On simulated data center power, 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 even when the operator tried to hide them.
A new paper on arXiv, "Detecting AI Training in Datacentre Power" by Tom Kimpson, Mauricio Baker and Emlyn Graham, addresses a verification problem in proposals for AI governance. A monitor has to tell whether a cluster is running a large t…
Continue the AI Intelligence Brief
Track frontier labs, chips, export controls, model releases, regulation, and AI infrastructure.
- 5 AI intelligence signals a day
- Frontier labs, compute, and chips
- Model releases and AI infrastructure
- Source-grounded analysis with confidence labels
The Global Intelligence Brief stays free.
Part of a tracked trend
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.
Townsquare
Start a discussion in Townsquare.
More from this edition
- Anthropic Says Claude Formally Verified 123 Open-Source Libraries at the Assembly Level
- Major US Banks Block or Restrict Outside AI Agents That Log In With Customers' Credentials
- OpenAI's GPT-6 Models Enter Private Preview on Google Cloud, Completing Its Spread Across the Three Largest US Clouds
- PHP Ships Emergency Patches for Two Critical Remote-Code-Execution Flaws That an OpenAI Audit Helped Find
- Microsoft Researchers Prove the 31-Year-Old KLS Conjecture With Human-Directed AI Agents
- Nvidia Pitches Europe on Sovereign AI Factories in Berlin, Citing Nearly $10 Billion Invested in a Year
- Poland's NASK Gains Access to OpenAI's Restricted GPT-5.5 Cyber Model Through a Government Partnership
Comments
0No comments yet.