Over five matched evaluation rounds, SkillSonar reduces ASR across all individual risk families, achieving reductions of 50.0 pp on Resource & Reliability, 42.9 pp on Execution Safety, 35.3 pp on Specification Integrity, 18.3 pp on Operational Side Effects, 61.5 pp on Privacy & Data Flow (OOD), and 44.8 pp on Capability Control (OOD).
来源:source-paper:Table 10, page 22
报告指标与观测值
attack_success_rate
m-t10-resource-reliability-noguard-asr
论文报告 61.3 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-resource-reliability-skillsonar-asr
论文报告 11.3 percentage_points
实际观测 — percentage_points
attack_success_rate_reduction
m-t10-resource-reliability-reduction
论文报告 50 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-execution-safety-noguard-asr
论文报告 48.6 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-execution-safety-skillsonar-asr
论文报告 5.7 percentage_points
实际观测 — percentage_points
attack_success_rate_reduction
m-t10-execution-safety-reduction
论文报告 42.9 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-specification-integrity-noguard-asr
论文报告 41.2 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-specification-integrity-skillsonar-asr
论文报告 5.9 percentage_points
实际观测 — percentage_points
attack_success_rate_reduction
m-t10-specification-integrity-reduction
论文报告 35.3 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-operational-side-effects-noguard-asr
论文报告 36.7 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-operational-side-effects-skillsonar-asr
论文报告 18.3 percentage_points
实际观测 — percentage_points
attack_success_rate_reduction
m-t10-operational-side-effects-reduction
论文报告 18.3 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-privacy-data-flow-noguard-asr
论文报告 67.7 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-privacy-data-flow-skillsonar-asr
论文报告 6.2 percentage_points
实际观测 — percentage_points
attack_success_rate_reduction
m-t10-privacy-data-flow-reduction
论文报告 61.5 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-capability-control-noguard-asr
论文报告 59.4 percentage_points
实际观测 — percentage_points
attack_success_rate
m-t10-capability-control-skillsonar-asr
论文报告 14.5 percentage_points
实际观测 — percentage_points
attack_success_rate_reduction
m-t10-capability-control-reduction
论文报告 44.8 percentage_points
实际观测 — percentage_points
Assessment(0)
这条 Claim 暂无已发布的不可变 Assessment。
关联运行(0)
这条 Claim 暂无关联执行。