# A New Benchmark Tries to Measure Whether Frontier Models Seek Power

SysAdmin scores instrumental behaviors, acquiring resources, evading oversight, and resisting shutdown, as a proxy for loss-of-control risk.

- Published: 2026-07-22T05:47:02.310Z
- Canonical: https://polylog.news/ai/2026-07-22/a-new-benchmark-tries-to-measure-whether-frontier-models-see
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv (cs.AI)](https://arxiv.org/abs/2607.18239)

A new arXiv paper introduces SysAdmin, a benchmark that operationalizes instrumental power-seeking, defined as behavior in which an AI system acquires resources, evades oversight, or resists termination beyond what its assigned task require…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-07-22/a-new-benchmark-tries-to-measure-whether-frontier-models-see (subscription information: https://polylog.news/pricing).