A guardrail classifier is reported as if unsafe were a property of text. It is not: it is a property of a policy, and the policy enters the model through its labels. We exploit an accident of an indus
1 comments
Policy Is Not Capability: Some Safety Benchmarks Measure Which Definition of "Unsafe" You Trained On was first seen on Hugging Face · Daily papers on Oct 20, 2026. Since then it appeared in 6 hourly checks across 2 days, peaking at #18 on Oct 21, 03:18 UTC. It was last on the list on Oct 21, 03:18 UTC.
Position on Hugging Face · Daily papers at every check (lower is better).
Best rank per day (UTC).
Platform metric as reported by Hugging Face.
Show that this trended, linking back to this page.
<a href="https://trendoday.com/hugging-face/papers/2610.25060-policy-is-not-capability-some-safety-benchmarks-measure"><img src="https://trendoday.com/badge/hugging-face/papers/2610.25060-policy-is-not-capability-some-safety-benchmarks-measure.svg" alt="Policy Is Not Capability: Some Safety Benchmarks Measure Which Definition of "Unsafe" You Trained On on TrendBoard" height="44"></a>
[](https://trendoday.com/hugging-face/papers/2610.25060-policy-is-not-capability-some-safety-benchmarks-measure)