“Energon”: Unveiling Transformers from GPU Power and Thermal Side-Channels
Arunava Chaudhuri 1 , Shubhi Shukla 2 , Sarani Bhattacharya 1 , and Debdeep Mukhopadhyay 1 · 2025
Read “Energon”: Unveiling Transformers from GPU Power and Thermal Side-Channels by Arunava Chaudhuri 1 , Shubhi Shukla 2 , Sarani Bhattacharya 1 , and Debdeep Mukhopadhyay 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-01768.txt.
Description
Four researchers show that a graphics chip's power draw and heat can betray the design of AI models running on it, in a 2025 paper. Transformer models are valuable property, and they increasingly run on shared cloud hardware where rivals rent neighboring space. The team's attack, Energon, watches the chip's energy and temperature wiggles from ordinary user access. From those signals it infers layer counts and attention head numbers for language and vision models.
In tests the attack named the right model family over eighty-nine percent of the time and the exact settings every time. It worked even while other jobs shared the chip. Worse, the stolen design details powered follow-up attacks that fooled the models over ninety-three percent of the time. The authors warn that cloud operators must treat heat and power as leaks, not noise.
Catalog record
| CREATOR | Arunava Chaudhuri 1 , Shubhi Shukla 2 , Sarani Bhattacharya 1 , and Debdeep Mukhopadhyay 1 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-01768