Benchmark RadarAI BENCHMARK PROFILE
AntiSkillBench
AntiSkillBench is a benchmark for evaluating privacy leakage and impersonation risks in persona-skills. It includes 7,500 persona-grounded dialogue traces and evaluates across three skill-distillation strategies.
- Released
- 2026-08-04
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Provides an end-to-end evaluation of risks and defenses in the persona-skill pipeline, enabling comparison of model safety across different settings.
Motivation
Persona skills distill personal interaction histories into portable and executable artifacts for downstream agents.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.