Attack success rate
A quantitative metric in AI safety evaluation measuring the proportion of successful attacks, defined as the number of successful attack samples divided by the total number of attack samples. In prompt injection evaluation, a lower ASR is generally regarded as safer.
Timeline (last 90 days)
Random error can be averaged out by increasing sample size, but systematic bias is amplified with sample size, causing attack success rates to be structurally overestimated or underestimated.
Attack success rate is not an intrinsic property of the target model but a joint product of the target model and the grader.
All Facts (2)
Random error can be averaged out by increasing sample size, but systematic bias is amplified with sample size, causing attack success rates to be structurally overestimated or underestimated.
50%UnverifiedAttack success rate is not an intrinsic property of the target model but a joint product of the target model and the grader.
50%