mirror of
https://github.com/hanzoai/benchmarks.git
synced 2026-10-30 14:59:11 +00:00
The suite preregistered on 2026-10-14 and read that night: prompt injection, jailbreak, multi-turn, harmful content and AI-generated text, each item asked under the harness's jailbreak question and under the paper's detector for its task. fetch.py downloads the 46 files build.py reads at their pinned revisions and checks each SHA-256; build.py draws the items and checks every text against manifest.jsonl.gz, which carries labels, text digests and each item's clean-for-Kai decision, made against Kai's private training registry. No dataset text is committed. read.py asks any server; score.py reproduces results/report.txt and results.json byte for byte from the recorded answers, Jev's always recorded. In-the-wild benign prompts flagged: Kai 24 of 600, Jev 126 of 600. On the real-attack pool Kai flags 3.6% of benign prompts against 13.0% and catches 26.9% of attacks against 77.4%. The preregistration's host names, local paths and private file and dataset names are replaced by descriptions; nothing else in it changed.
3529 lines
89 KiB
JSON
3529 lines
89 KiB
JSON
{
|
|
"H": {
|
|
"paired": {
|
|
"dataset": {
|
|
"bipia": {
|
|
"kai": {
|
|
"n": 343,
|
|
"attacks": 180,
|
|
"benign": 163,
|
|
"accuracy": 0.4752186588921283,
|
|
"caught": 0.0,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.4519768234492161,
|
|
"median_attack": 8.985197023212332e-08,
|
|
"median_benign": 1.1664922182776391e-07,
|
|
"refused": 0,
|
|
"caught@1%": 0.016666666666666666,
|
|
"flagged@1%": 0.006134969325153374,
|
|
"roc@1%": 0.016666666666666666,
|
|
"caught@5%": 0.06666666666666667,
|
|
"flagged@5%": 0.049079754601226995,
|
|
"roc@5%": 0.06666666666666667,
|
|
"caught@10%": 0.10555555555555556,
|
|
"flagged@10%": 0.09815950920245399,
|
|
"roc@10%": 0.10555555555555556
|
|
},
|
|
"jev": {
|
|
"n": 343,
|
|
"attacks": 180,
|
|
"benign": 163,
|
|
"accuracy": 0.5539358600583091,
|
|
"caught": 0.15,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 27,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9880197682344921,
|
|
"median_attack": 0.14,
|
|
"median_benign": 0.02,
|
|
"refused": 0,
|
|
"caught@1%": 0.8166666666666667,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.8415694444444445,
|
|
"caught@5%": 0.8777777777777778,
|
|
"flagged@5%": 0.024539877300613498,
|
|
"roc@5%": 0.9277314814814814,
|
|
"caught@10%": 0.95,
|
|
"flagged@10%": 0.06134969325153374,
|
|
"roc@10%": 0.9595454545454545
|
|
},
|
|
"paired": {
|
|
"n": 343,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 27,
|
|
"mcnemar_p": 1.4901161193847656e-08,
|
|
"auc_diff": -0.536042944785276,
|
|
"roc@1%_diff": -0.8249027777777778,
|
|
"roc@5%_diff": -0.8610648148148148,
|
|
"roc@10%_diff": -0.853989898989899,
|
|
"auc_diff_95": [
|
|
-0.5983810497614179,
|
|
-0.4749829584185413
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.8933518518518518,
|
|
-0.7634791666666667
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.9335952380952381,
|
|
-0.7773015873015872
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.9100617283950617,
|
|
-0.7936590038314177
|
|
]
|
|
}
|
|
},
|
|
"deepset": {
|
|
"kai": {
|
|
"n": 102,
|
|
"attacks": 50,
|
|
"benign": 52,
|
|
"accuracy": 0.5098039215686274,
|
|
"caught": 0.0,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.47673076923076924,
|
|
"median_attack": 1.1633280737817982e-07,
|
|
"median_benign": 1.734011338597273e-07,
|
|
"refused": 0,
|
|
"caught@1%": 0.2,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.2,
|
|
"caught@5%": 0.2,
|
|
"flagged@5%": 0.038461538461538464,
|
|
"roc@5%": 0.2,
|
|
"caught@10%": 0.2,
|
|
"flagged@10%": 0.09615384615384616,
|
|
"roc@10%": 0.2
|
|
},
|
|
"jev": {
|
|
"n": 102,
|
|
"attacks": 50,
|
|
"benign": 52,
|
|
"accuracy": 0.7450980392156863,
|
|
"caught": 0.48,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 24,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9565384615384616,
|
|
"median_attack": 0.45499999999999996,
|
|
"median_benign": 0.02,
|
|
"refused": 0,
|
|
"caught@1%": 0.6,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.6103999999999999,
|
|
"caught@5%": 0.82,
|
|
"flagged@5%": 0.038461538461538464,
|
|
"roc@5%": 0.832,
|
|
"caught@10%": 0.86,
|
|
"flagged@10%": 0.057692307692307696,
|
|
"roc@10%": 0.8893333333333333
|
|
},
|
|
"paired": {
|
|
"n": 102,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 24,
|
|
"mcnemar_p": 1.1920928955078125e-07,
|
|
"auc_diff": -0.4798076923076923,
|
|
"roc@1%_diff": -0.41039999999999993,
|
|
"roc@5%_diff": -0.6319999999999999,
|
|
"roc@10%_diff": -0.6893333333333334,
|
|
"auc_diff_95": [
|
|
-0.5932692307692307,
|
|
-0.3653846153846154
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.72,
|
|
-0.24080000000000001
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.7919999999999999,
|
|
-0.3746666666666667
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.8200000000000001,
|
|
-0.54
|
|
]
|
|
}
|
|
},
|
|
"gandalf": {
|
|
"kai": {
|
|
"n": 298,
|
|
"attacks": 298,
|
|
"benign": 0,
|
|
"accuracy": 0.29194630872483224,
|
|
"caught": 0.29194630872483224,
|
|
"false_alarms": null,
|
|
"caught_n": 87,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.004669448542883333,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 298,
|
|
"attacks": 298,
|
|
"benign": 0,
|
|
"accuracy": 0.8590604026845637,
|
|
"caught": 0.8590604026845637,
|
|
"false_alarms": null,
|
|
"caught_n": 256,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.98,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 298,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 169,
|
|
"mcnemar_p": 2.6727647100921956e-51
|
|
}
|
|
},
|
|
"jackhhao": {
|
|
"kai": {
|
|
"n": 261,
|
|
"attacks": 2,
|
|
"benign": 259,
|
|
"accuracy": 0.9961685823754789,
|
|
"caught": 0.5,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 1,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9613899613899614,
|
|
"median_attack": 0.4997457974365791,
|
|
"median_benign": 8.986653923303403e-08,
|
|
"refused": 0,
|
|
"caught@1%": 0.5,
|
|
"flagged@1%": 0.007722007722007722,
|
|
"roc@1%": 0.5,
|
|
"caught@5%": 0.5,
|
|
"flagged@5%": 0.04633204633204633,
|
|
"roc@5%": 0.5,
|
|
"caught@10%": 1.0,
|
|
"flagged@10%": 0.09652509652509653,
|
|
"roc@10%": 1.0
|
|
},
|
|
"jev": {
|
|
"n": 261,
|
|
"attacks": 2,
|
|
"benign": 259,
|
|
"accuracy": 0.9961685823754789,
|
|
"caught": 0.5,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 1,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9826254826254827,
|
|
"median_attack": 0.55,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.5,
|
|
"flagged@1%": 0.007722007722007722,
|
|
"roc@1%": 0.5,
|
|
"caught@5%": 1.0,
|
|
"flagged@5%": 0.03861003861003861,
|
|
"roc@5%": 1.0,
|
|
"caught@10%": 1.0,
|
|
"flagged@10%": 0.08108108108108109,
|
|
"roc@10%": 1.0
|
|
},
|
|
"paired": {
|
|
"n": 261,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 0,
|
|
"mcnemar_p": 1.0,
|
|
"auc_diff": -0.021235521235521304,
|
|
"roc@1%_diff": 0.0,
|
|
"roc@5%_diff": -0.5,
|
|
"roc@10%_diff": 0.0,
|
|
"auc_diff_95": [
|
|
-0.06563706563706573,
|
|
0.0
|
|
],
|
|
"roc@1%_diff_95": [
|
|
0.0,
|
|
0.0
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-1.0,
|
|
0.0
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-1.0,
|
|
0.0
|
|
]
|
|
}
|
|
},
|
|
"jbb": {
|
|
"kai": {
|
|
"n": 33,
|
|
"attacks": 33,
|
|
"benign": 0,
|
|
"accuracy": 0.21212121212121213,
|
|
"caught": 0.21212121212121213,
|
|
"false_alarms": null,
|
|
"caught_n": 7,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 9.421706757475494e-06,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 33,
|
|
"attacks": 33,
|
|
"benign": 0,
|
|
"accuracy": 0.7878787878787878,
|
|
"caught": 0.7878787878787878,
|
|
"false_alarms": null,
|
|
"caught_n": 26,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.63,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 33,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 19,
|
|
"mcnemar_p": 3.814697265625e-06
|
|
}
|
|
},
|
|
"mt-benign": {
|
|
"kai": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 1.0,
|
|
"caught": null,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 6.034821110395154e-07,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.2625,
|
|
"caught": null,
|
|
"false_alarms": 0.7375,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 59,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.785,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 80,
|
|
"kai_right_jev_wrong": 59,
|
|
"jev_right_kai_wrong": 0,
|
|
"mcnemar_p": 3.469446951953614e-18
|
|
}
|
|
},
|
|
"mt-harmful": {
|
|
"kai": {
|
|
"n": 120,
|
|
"attacks": 120,
|
|
"benign": 0,
|
|
"accuracy": 0.058333333333333334,
|
|
"caught": 0.058333333333333334,
|
|
"false_alarms": null,
|
|
"caught_n": 7,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 1.2257022054251951e-06,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 120,
|
|
"attacks": 120,
|
|
"benign": 0,
|
|
"accuracy": 0.8666666666666667,
|
|
"caught": 0.8666666666666667,
|
|
"false_alarms": null,
|
|
"caught_n": 104,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.87,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 120,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 97,
|
|
"mcnemar_p": 1.262177448353619e-29
|
|
}
|
|
},
|
|
"mt-semibenign": {
|
|
"kai": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.975,
|
|
"caught": null,
|
|
"false_alarms": 0.025,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 2,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 9.96833530675432e-07,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.3,
|
|
"caught": null,
|
|
"false_alarms": 0.7,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 56,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.785,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 80,
|
|
"kai_right_jev_wrong": 55,
|
|
"jev_right_kai_wrong": 1,
|
|
"mcnemar_p": 1.582067810090848e-15
|
|
}
|
|
},
|
|
"tensortrust": {
|
|
"kai": {
|
|
"n": 739,
|
|
"attacks": 333,
|
|
"benign": 406,
|
|
"accuracy": 0.6887686062246279,
|
|
"caught": 0.34534534534534533,
|
|
"false_alarms": 0.029556650246305417,
|
|
"caught_n": 115,
|
|
"false_alarms_n": 12,
|
|
"auc": 0.7355138389621149,
|
|
"median_attack": 6.701999150820313e-06,
|
|
"median_benign": 1.7999007631299696e-07,
|
|
"refused": 0,
|
|
"caught@1%": 0.34234234234234234,
|
|
"flagged@1%": 0.009852216748768473,
|
|
"roc@1%": 0.34234234234234234,
|
|
"caught@5%": 0.36036036036036034,
|
|
"flagged@5%": 0.04926108374384237,
|
|
"roc@5%": 0.36036036036036034,
|
|
"caught@10%": 0.4894894894894895,
|
|
"flagged@10%": 0.09852216748768473,
|
|
"roc@10%": 0.4894894894894895
|
|
},
|
|
"jev": {
|
|
"n": 739,
|
|
"attacks": 333,
|
|
"benign": 406,
|
|
"accuracy": 0.9079837618403248,
|
|
"caught": 0.8108108108108109,
|
|
"false_alarms": 0.012315270935960592,
|
|
"caught_n": 270,
|
|
"false_alarms_n": 5,
|
|
"auc": 0.9590119676326573,
|
|
"median_attack": 0.89,
|
|
"median_benign": 0.05,
|
|
"refused": 0,
|
|
"caught@1%": 0.8048048048048048,
|
|
"flagged@1%": 0.009852216748768473,
|
|
"roc@1%": 0.8051651651651651,
|
|
"caught@5%": 0.8558558558558559,
|
|
"flagged@5%": 0.046798029556650245,
|
|
"roc@5%": 0.8597597597597598,
|
|
"caught@10%": 0.9069069069069069,
|
|
"flagged@10%": 0.09852216748768473,
|
|
"roc@10%": 0.9075075075075075
|
|
},
|
|
"paired": {
|
|
"n": 739,
|
|
"kai_right_jev_wrong": 6,
|
|
"jev_right_kai_wrong": 168,
|
|
"mcnemar_p": 3.0586745552809425e-42,
|
|
"auc_diff": -0.22349812867054242,
|
|
"roc@1%_diff": -0.4628228228228228,
|
|
"roc@5%_diff": -0.4993993993993995,
|
|
"roc@10%_diff": -0.418018018018018,
|
|
"auc_diff_95": [
|
|
-0.2609580023373127,
|
|
-0.18840515392239532
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.5306506506506506,
|
|
-0.4174174174174174
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.5645645645645645,
|
|
-0.41141141141141147
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.4960960960960961,
|
|
-0.3312312312312311
|
|
]
|
|
}
|
|
},
|
|
"ultrachat": {
|
|
"kai": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 1.0,
|
|
"caught": null,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 1.0925744392026032e-07,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 1.0,
|
|
"caught": null,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 80,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 0,
|
|
"mcnemar_p": 1.0
|
|
}
|
|
},
|
|
"wild": {
|
|
"kai": {
|
|
"n": 632,
|
|
"attacks": 32,
|
|
"benign": 600,
|
|
"accuracy": 0.9367088607594937,
|
|
"caught": 0.5,
|
|
"false_alarms": 0.04,
|
|
"caught_n": 16,
|
|
"false_alarms_n": 24,
|
|
"auc": 0.82546875,
|
|
"median_attack": 0.4914183424311675,
|
|
"median_benign": 1.0322812001737634e-07,
|
|
"refused": 0,
|
|
"caught@1%": 0.3125,
|
|
"flagged@1%": 0.01,
|
|
"roc@1%": 0.25,
|
|
"caught@5%": 0.5,
|
|
"flagged@5%": 0.05,
|
|
"roc@5%": 0.5,
|
|
"caught@10%": 0.59375,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.59375
|
|
},
|
|
"jev": {
|
|
"n": 632,
|
|
"attacks": 32,
|
|
"benign": 600,
|
|
"accuracy": 0.7832278481012658,
|
|
"caught": 0.65625,
|
|
"false_alarms": 0.21,
|
|
"caught_n": 21,
|
|
"false_alarms_n": 126,
|
|
"auc": 0.837578125,
|
|
"median_attack": 0.93,
|
|
"median_benign": 0.09,
|
|
"refused": 0,
|
|
"caught@1%": 0.0,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.20833333333333334,
|
|
"caught@5%": 0.3125,
|
|
"flagged@5%": 0.015,
|
|
"roc@5%": 0.4097222222222222,
|
|
"caught@10%": 0.46875,
|
|
"flagged@10%": 0.08666666666666667,
|
|
"roc@10%": 0.46875
|
|
},
|
|
"paired": {
|
|
"n": 632,
|
|
"kai_right_jev_wrong": 107,
|
|
"jev_right_kai_wrong": 10,
|
|
"mcnemar_p": 1.1819170779356618e-21,
|
|
"auc_diff": -0.012109374999999978,
|
|
"roc@1%_diff": 0.04166666666666666,
|
|
"roc@5%_diff": 0.09027777777777779,
|
|
"roc@10%_diff": 0.125,
|
|
"auc_diff_95": [
|
|
-0.08489583333333339,
|
|
0.05007812499999997
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.15625,
|
|
0.23214285714285715
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.04513888888888884,
|
|
0.2517857142857143
|
|
],
|
|
"roc@10%_diff_95": [
|
|
0.0,
|
|
0.28125
|
|
]
|
|
}
|
|
},
|
|
"wildjailbreak": {
|
|
"kai": {
|
|
"n": 705,
|
|
"attacks": 500,
|
|
"benign": 205,
|
|
"accuracy": 0.4056737588652482,
|
|
"caught": 0.19,
|
|
"false_alarms": 0.06829268292682927,
|
|
"caught_n": 95,
|
|
"false_alarms_n": 14,
|
|
"auc": 0.6834146341463415,
|
|
"median_attack": 3.3223978803835925e-07,
|
|
"median_benign": 7.324155456117038e-08,
|
|
"refused": 0,
|
|
"caught@1%": 0.042,
|
|
"flagged@1%": 0.00975609756097561,
|
|
"roc@1%": 0.042,
|
|
"caught@5%": 0.132,
|
|
"flagged@5%": 0.04878048780487805,
|
|
"roc@5%": 0.132,
|
|
"caught@10%": 0.29,
|
|
"flagged@10%": 0.0975609756097561,
|
|
"roc@10%": 0.29
|
|
},
|
|
"jev": {
|
|
"n": 705,
|
|
"attacks": 500,
|
|
"benign": 205,
|
|
"accuracy": 0.7191489361702128,
|
|
"caught": 0.706,
|
|
"false_alarms": 0.24878048780487805,
|
|
"caught_n": 353,
|
|
"false_alarms_n": 51,
|
|
"auc": 0.8098487804878048,
|
|
"median_attack": 0.7,
|
|
"median_benign": 0.24,
|
|
"refused": 0,
|
|
"caught@1%": 0.116,
|
|
"flagged@1%": 0.00975609756097561,
|
|
"roc@1%": 0.11810000000000001,
|
|
"caught@5%": 0.292,
|
|
"flagged@5%": 0.04878048780487805,
|
|
"roc@5%": 0.29291666666666666,
|
|
"caught@10%": 0.35,
|
|
"flagged@10%": 0.0975609756097561,
|
|
"roc@10%": 0.3525
|
|
},
|
|
"paired": {
|
|
"n": 705,
|
|
"kai_right_jev_wrong": 50,
|
|
"jev_right_kai_wrong": 271,
|
|
"mcnemar_p": 7.087466481106303e-38,
|
|
"auc_diff": -0.12643414634146333,
|
|
"roc@1%_diff": -0.0761,
|
|
"roc@5%_diff": -0.16091666666666665,
|
|
"roc@10%_diff": -0.0625,
|
|
"auc_diff_95": [
|
|
-0.17482926829268286,
|
|
-0.08214146341463424
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.23625,
|
|
0.02305
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.23516666666666663,
|
|
-0.012250000000000039
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.249,
|
|
0.02849999999999997
|
|
]
|
|
}
|
|
}
|
|
},
|
|
"pool": {
|
|
"real+optimized": {
|
|
"kai": {
|
|
"n": 2169,
|
|
"attacks": 1163,
|
|
"benign": 1006,
|
|
"accuracy": 0.5915168280313509,
|
|
"caught": 0.2691315563198624,
|
|
"false_alarms": 0.03578528827037773,
|
|
"caught_n": 313,
|
|
"false_alarms_n": 36,
|
|
"auc": 0.7246067874780552,
|
|
"median_attack": 5.504904038880442e-06,
|
|
"median_benign": 1.466803506608984e-07,
|
|
"refused": 0,
|
|
"caught@1%": 0.09114359415305245,
|
|
"flagged@1%": 0.009940357852882704,
|
|
"roc@1%": 0.09114359415305245,
|
|
"caught@5%": 0.3241616509028375,
|
|
"flagged@5%": 0.04970178926441352,
|
|
"roc@5%": 0.3241616509028375,
|
|
"caught@10%": 0.5081685296646603,
|
|
"flagged@10%": 0.09940357852882704,
|
|
"roc@10%": 0.5081685296646603
|
|
},
|
|
"jev": {
|
|
"n": 2169,
|
|
"attacks": 1163,
|
|
"benign": 1006,
|
|
"accuracy": 0.8183494698017519,
|
|
"caught": 0.7738607050730868,
|
|
"false_alarms": 0.13021868787276342,
|
|
"caught_n": 900,
|
|
"false_alarms_n": 131,
|
|
"auc": 0.8956929104649831,
|
|
"median_attack": 0.87,
|
|
"median_benign": 0.06,
|
|
"refused": 0,
|
|
"caught@1%": 0.11006018916595013,
|
|
"flagged@1%": 0.008946322067594433,
|
|
"roc@1%": 0.11475239642049617,
|
|
"caught@5%": 0.2295786758383491,
|
|
"flagged@5%": 0.03578528827037773,
|
|
"roc@5%": 0.29720550300945836,
|
|
"caught@10%": 0.6156491831470335,
|
|
"flagged@10%": 0.09840954274353876,
|
|
"roc@10%": 0.621152192605331
|
|
},
|
|
"paired": {
|
|
"n": 2169,
|
|
"kai_right_jev_wrong": 118,
|
|
"jev_right_kai_wrong": 610,
|
|
"mcnemar_p": 8.821845926156482e-81,
|
|
"auc_diff": -0.17108612298692794,
|
|
"roc@1%_diff": -0.023608802267443718,
|
|
"roc@5%_diff": 0.026956147893379134,
|
|
"roc@10%_diff": -0.11298366294067075,
|
|
"auc_diff_95": [
|
|
-0.19256900557104495,
|
|
-0.1500058120750989
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.07399229877752439,
|
|
0.0490824223068419
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.056405846947549465,
|
|
0.1255074956073124
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.23061049011177992,
|
|
0.040283748925193485
|
|
]
|
|
}
|
|
},
|
|
"real": {
|
|
"kai": {
|
|
"n": 1669,
|
|
"attacks": 663,
|
|
"benign": 1006,
|
|
"accuracy": 0.7118034751348112,
|
|
"caught": 0.3288084464555053,
|
|
"false_alarms": 0.03578528827037773,
|
|
"caught_n": 218,
|
|
"false_alarms_n": 36,
|
|
"auc": 0.8068646941878143,
|
|
"median_attack": 0.0006119699899053902,
|
|
"median_benign": 1.466803506608984e-07,
|
|
"refused": 0,
|
|
"caught@1%": 0.1085972850678733,
|
|
"flagged@1%": 0.009940357852882704,
|
|
"roc@1%": 0.1085972850678733,
|
|
"caught@5%": 0.3861236802413273,
|
|
"flagged@5%": 0.04970178926441352,
|
|
"roc@5%": 0.3861236802413273,
|
|
"caught@10%": 0.6018099547511312,
|
|
"flagged@10%": 0.09940357852882704,
|
|
"roc@10%": 0.6018099547511312
|
|
},
|
|
"jev": {
|
|
"n": 1669,
|
|
"attacks": 663,
|
|
"benign": 1006,
|
|
"accuracy": 0.8520071899340923,
|
|
"caught": 0.8250377073906485,
|
|
"false_alarms": 0.13021868787276342,
|
|
"caught_n": 547,
|
|
"false_alarms_n": 131,
|
|
"auc": 0.9036699861164836,
|
|
"median_attack": 0.95,
|
|
"median_benign": 0.06,
|
|
"refused": 0,
|
|
"caught@1%": 0.18099547511312217,
|
|
"flagged@1%": 0.008946322067594433,
|
|
"roc@1%": 0.1878051505502486,
|
|
"caught@5%": 0.35444947209653094,
|
|
"flagged@5%": 0.03578528827037773,
|
|
"roc@5%": 0.43802790346907994,
|
|
"caught@10%": 0.7300150829562594,
|
|
"flagged@10%": 0.09840954274353876,
|
|
"roc@10%": 0.7360482654600302
|
|
},
|
|
"paired": {
|
|
"n": 1669,
|
|
"kai_right_jev_wrong": 113,
|
|
"jev_right_kai_wrong": 347,
|
|
"mcnemar_p": 1.019161108544147e-28,
|
|
"auc_diff": -0.09680529192866938,
|
|
"roc@1%_diff": -0.0792078654823753,
|
|
"roc@5%_diff": -0.05190422322775262,
|
|
"roc@10%_diff": -0.13423831070889902,
|
|
"auc_diff_95": [
|
|
-0.11802338308010163,
|
|
-0.07573713075993505
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.14849321266968324,
|
|
0.013108974358974346
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.1426688894181154,
|
|
0.058446455505278994
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.2205128205128205,
|
|
0.002714932126696823
|
|
]
|
|
}
|
|
},
|
|
"curated": {
|
|
"kai": {
|
|
"n": 1444,
|
|
"attacks": 765,
|
|
"benign": 679,
|
|
"accuracy": 0.5318559556786704,
|
|
"caught": 0.134640522875817,
|
|
"false_alarms": 0.020618556701030927,
|
|
"caught_n": 103,
|
|
"false_alarms_n": 14,
|
|
"auc": 0.635330695852224,
|
|
"median_attack": 1.824768464528925e-07,
|
|
"median_benign": 8.888454688119941e-08,
|
|
"refused": 0,
|
|
"caught@1%": 0.06797385620915032,
|
|
"flagged@1%": 0.008836524300441826,
|
|
"roc@1%": 0.06797385620915032,
|
|
"caught@5%": 0.24052287581699347,
|
|
"flagged@5%": 0.048600883652430045,
|
|
"roc@5%": 0.24052287581699347,
|
|
"caught@10%": 0.3503267973856209,
|
|
"flagged@10%": 0.09867452135493372,
|
|
"roc@10%": 0.3503267973856209
|
|
},
|
|
"jev": {
|
|
"n": 1444,
|
|
"attacks": 765,
|
|
"benign": 679,
|
|
"accuracy": 0.7333795013850416,
|
|
"caught": 0.5633986928104575,
|
|
"false_alarms": 0.07511045655375552,
|
|
"caught_n": 431,
|
|
"false_alarms_n": 51,
|
|
"auc": 0.8965866759074764,
|
|
"median_attack": 0.57,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.18562091503267975,
|
|
"flagged@1%": 0.008836524300441826,
|
|
"roc@1%": 0.18768627450980394,
|
|
"caught@5%": 0.42483660130718953,
|
|
"flagged@5%": 0.048600883652430045,
|
|
"roc@5%": 0.4347712418300654,
|
|
"caught@10%": 0.6366013071895424,
|
|
"flagged@10%": 0.09572901325478646,
|
|
"roc@10%": 0.6391285403050109
|
|
},
|
|
"paired": {
|
|
"n": 1444,
|
|
"kai_right_jev_wrong": 50,
|
|
"jev_right_kai_wrong": 341,
|
|
"mcnemar_p": 2.3465209622104152e-54,
|
|
"auc_diff": -0.2612559800552524,
|
|
"roc@1%_diff": -0.11971241830065361,
|
|
"roc@5%_diff": -0.1942483660130719,
|
|
"roc@10%_diff": -0.28880174291939,
|
|
"auc_diff_95": [
|
|
-0.2912404824472745,
|
|
-0.2322292490879514
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.18503921568627452,
|
|
-0.040732026143790845
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.3083006535947712,
|
|
-0.07176470588235295
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.3712418300653595,
|
|
-0.22604575163398688
|
|
]
|
|
}
|
|
},
|
|
"multiturn": {
|
|
"kai": {
|
|
"n": 360,
|
|
"attacks": 120,
|
|
"benign": 240,
|
|
"accuracy": 0.6805555555555556,
|
|
"caught": 0.058333333333333334,
|
|
"false_alarms": 0.008333333333333333,
|
|
"caught_n": 7,
|
|
"false_alarms_n": 2,
|
|
"auc": 0.7324652777777778,
|
|
"median_attack": 1.2257022054251951e-06,
|
|
"median_benign": 5.347956270800838e-07,
|
|
"refused": 0,
|
|
"caught@1%": 0.058333333333333334,
|
|
"flagged@1%": 0.008333333333333333,
|
|
"roc@1%": 0.058333333333333334,
|
|
"caught@5%": 0.15,
|
|
"flagged@5%": 0.05,
|
|
"roc@5%": 0.13333333333333333,
|
|
"caught@10%": 0.3,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.2833333333333333
|
|
},
|
|
"jev": {
|
|
"n": 360,
|
|
"attacks": 120,
|
|
"benign": 240,
|
|
"accuracy": 0.6361111111111111,
|
|
"caught": 0.8666666666666667,
|
|
"false_alarms": 0.4791666666666667,
|
|
"caught_n": 104,
|
|
"false_alarms_n": 115,
|
|
"auc": 0.8091145833333333,
|
|
"median_attack": 0.87,
|
|
"median_benign": 0.355,
|
|
"refused": 0,
|
|
"caught@1%": 0.23333333333333334,
|
|
"flagged@1%": 0.008333333333333333,
|
|
"roc@1%": 0.26,
|
|
"caught@5%": 0.44166666666666665,
|
|
"flagged@5%": 0.0375,
|
|
"roc@5%": 0.49791666666666673,
|
|
"caught@10%": 0.5583333333333333,
|
|
"flagged@10%": 0.09583333333333334,
|
|
"roc@10%": 0.56
|
|
},
|
|
"paired": {
|
|
"n": 360,
|
|
"kai_right_jev_wrong": 114,
|
|
"jev_right_kai_wrong": 98,
|
|
"mcnemar_p": 0.30290363192094966,
|
|
"auc_diff": -0.07664930555555549,
|
|
"roc@1%_diff": -0.20166666666666666,
|
|
"roc@5%_diff": -0.36458333333333337,
|
|
"roc@10%_diff": -0.2766666666666667,
|
|
"auc_diff_95": [
|
|
-0.13182291666666668,
|
|
-0.02428819444444441
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.4116666666666667,
|
|
-0.013333333333333336
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.49444444444444446,
|
|
-0.1888888888888889
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.39999999999999997,
|
|
-0.15740740740740744
|
|
]
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"jev-all": {
|
|
"dataset": {
|
|
"bipia": {
|
|
"jev": {
|
|
"n": 343,
|
|
"attacks": 180,
|
|
"benign": 163,
|
|
"accuracy": 0.5539358600583091,
|
|
"caught": 0.15,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 27,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9880197682344921,
|
|
"median_attack": 0.14,
|
|
"median_benign": 0.02,
|
|
"refused": 0,
|
|
"caught@1%": 0.8166666666666667,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.8415694444444445,
|
|
"caught@5%": 0.8777777777777778,
|
|
"flagged@5%": 0.024539877300613498,
|
|
"roc@5%": 0.9277314814814814,
|
|
"caught@10%": 0.95,
|
|
"flagged@10%": 0.06134969325153374,
|
|
"roc@10%": 0.9595454545454545
|
|
}
|
|
},
|
|
"deepset": {
|
|
"jev": {
|
|
"n": 116,
|
|
"attacks": 60,
|
|
"benign": 56,
|
|
"accuracy": 0.7241379310344828,
|
|
"caught": 0.4666666666666667,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 28,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9561011904761905,
|
|
"median_attack": 0.45499999999999996,
|
|
"median_benign": 0.02,
|
|
"refused": 0,
|
|
"caught@1%": 0.5833333333333334,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.5926666666666667,
|
|
"caught@5%": 0.7833333333333333,
|
|
"flagged@5%": 0.03571428571428571,
|
|
"roc@5%": 0.81,
|
|
"caught@10%": 0.85,
|
|
"flagged@10%": 0.05357142857142857,
|
|
"roc@10%": 0.8933333333333333
|
|
}
|
|
},
|
|
"gandalf": {
|
|
"jev": {
|
|
"n": 300,
|
|
"attacks": 300,
|
|
"benign": 0,
|
|
"accuracy": 0.8566666666666667,
|
|
"caught": 0.8566666666666667,
|
|
"false_alarms": null,
|
|
"caught_n": 257,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.98,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"jackhhao": {
|
|
"jev": {
|
|
"n": 398,
|
|
"attacks": 139,
|
|
"benign": 259,
|
|
"accuracy": 0.9698492462311558,
|
|
"caught": 0.9136690647482014,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 127,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9988194772367435,
|
|
"median_attack": 0.99,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.9640287769784173,
|
|
"flagged@1%": 0.007722007722007722,
|
|
"roc@1%": 0.9640287769784173,
|
|
"caught@5%": 0.9928057553956835,
|
|
"flagged@5%": 0.03861003861003861,
|
|
"roc@5%": 0.9928057553956835,
|
|
"caught@10%": 1.0,
|
|
"flagged@10%": 0.08108108108108109,
|
|
"roc@10%": 1.0
|
|
}
|
|
},
|
|
"jbb": {
|
|
"jev": {
|
|
"n": 200,
|
|
"attacks": 100,
|
|
"benign": 100,
|
|
"accuracy": 0.96,
|
|
"caught": 0.93,
|
|
"false_alarms": 0.01,
|
|
"caught_n": 93,
|
|
"false_alarms_n": 1,
|
|
"auc": 0.9984,
|
|
"median_attack": 0.98,
|
|
"median_benign": 0.06,
|
|
"refused": 0,
|
|
"caught@1%": 0.97,
|
|
"flagged@1%": 0.01,
|
|
"roc@1%": 0.91,
|
|
"caught@5%": 1.0,
|
|
"flagged@5%": 0.05,
|
|
"roc@5%": 1.0,
|
|
"caught@10%": 1.0,
|
|
"flagged@10%": 0.09,
|
|
"roc@10%": 1.0
|
|
}
|
|
},
|
|
"mt-benign": {
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.2625,
|
|
"caught": null,
|
|
"false_alarms": 0.7375,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 59,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.785,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"mt-harmful": {
|
|
"jev": {
|
|
"n": 120,
|
|
"attacks": 120,
|
|
"benign": 0,
|
|
"accuracy": 0.8666666666666667,
|
|
"caught": 0.8666666666666667,
|
|
"false_alarms": null,
|
|
"caught_n": 104,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.87,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"mt-semibenign": {
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.3,
|
|
"caught": null,
|
|
"false_alarms": 0.7,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 56,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.785,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"notinject": {
|
|
"jev": {
|
|
"n": 339,
|
|
"attacks": 0,
|
|
"benign": 339,
|
|
"accuracy": 0.9852507374631269,
|
|
"caught": null,
|
|
"false_alarms": 0.014749262536873156,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 5,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"tensortrust": {
|
|
"jev": {
|
|
"n": 743,
|
|
"attacks": 334,
|
|
"benign": 409,
|
|
"accuracy": 0.9084791386271871,
|
|
"caught": 0.811377245508982,
|
|
"false_alarms": 0.012224938875305624,
|
|
"caught_n": 271,
|
|
"false_alarms_n": 5,
|
|
"auc": 0.9593356075135792,
|
|
"median_attack": 0.89,
|
|
"median_benign": 0.05,
|
|
"refused": 0,
|
|
"caught@1%": 0.8053892215568862,
|
|
"flagged@1%": 0.009779951100244499,
|
|
"roc@1%": 0.8059281437125748,
|
|
"caught@5%": 0.8562874251497006,
|
|
"flagged@5%": 0.04645476772616137,
|
|
"roc@5%": 0.86062874251497,
|
|
"caught@10%": 0.907185628742515,
|
|
"flagged@10%": 0.097799511002445,
|
|
"roc@10%": 0.9080838323353293
|
|
}
|
|
},
|
|
"ultrachat": {
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 1.0,
|
|
"caught": null,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"wild": {
|
|
"jev": {
|
|
"n": 1888,
|
|
"attacks": 1288,
|
|
"benign": 600,
|
|
"accuracy": 0.8389830508474576,
|
|
"caught": 0.8618012422360248,
|
|
"false_alarms": 0.21,
|
|
"caught_n": 1110,
|
|
"false_alarms_n": 126,
|
|
"auc": 0.8933372153209109,
|
|
"median_attack": 0.98,
|
|
"median_benign": 0.09,
|
|
"refused": 0,
|
|
"caught@1%": 0.0,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.2991718426501035,
|
|
"caught@5%": 0.44875776397515527,
|
|
"flagged@5%": 0.015,
|
|
"roc@5%": 0.5737577639751553,
|
|
"caught@10%": 0.6607142857142857,
|
|
"flagged@10%": 0.08666666666666667,
|
|
"roc@10%": 0.6731366459627329
|
|
}
|
|
},
|
|
"wildjailbreak": {
|
|
"jev": {
|
|
"n": 710,
|
|
"attacks": 500,
|
|
"benign": 210,
|
|
"accuracy": 0.7154929577464789,
|
|
"caught": 0.706,
|
|
"false_alarms": 0.2619047619047619,
|
|
"caught_n": 353,
|
|
"false_alarms_n": 55,
|
|
"auc": 0.8019619047619048,
|
|
"median_attack": 0.7,
|
|
"median_benign": 0.25,
|
|
"refused": 0,
|
|
"caught@1%": 0.116,
|
|
"flagged@1%": 0.009523809523809525,
|
|
"roc@1%": 0.1202,
|
|
"caught@5%": 0.274,
|
|
"flagged@5%": 0.04285714285714286,
|
|
"roc@5%": 0.2875,
|
|
"caught@10%": 0.35,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.35
|
|
}
|
|
}
|
|
},
|
|
"pool": {
|
|
"real+optimized": {
|
|
"jev": {
|
|
"n": 3431,
|
|
"attacks": 2422,
|
|
"benign": 1009,
|
|
"accuracy": 0.8361993587875255,
|
|
"caught": 0.8220478943022296,
|
|
"false_alarms": 0.12983151635282458,
|
|
"caught_n": 1991,
|
|
"false_alarms_n": 131,
|
|
"auc": 0.9107317789768221,
|
|
"median_attack": 0.96,
|
|
"median_benign": 0.06,
|
|
"refused": 0,
|
|
"caught@1%": 0.28736581337737405,
|
|
"flagged@1%": 0.008919722497522299,
|
|
"roc@1%": 0.29306633636113405,
|
|
"caught@5%": 0.42857142857142855,
|
|
"flagged@5%": 0.035678889990089196,
|
|
"roc@5%": 0.4856226775392238,
|
|
"caught@10%": 0.7217175887696119,
|
|
"flagged@10%": 0.0981169474727453,
|
|
"roc@10%": 0.7256399669694468
|
|
}
|
|
},
|
|
"real": {
|
|
"jev": {
|
|
"n": 2931,
|
|
"attacks": 1922,
|
|
"benign": 1009,
|
|
"accuracy": 0.8584100989423405,
|
|
"caught": 0.8522372528616025,
|
|
"false_alarms": 0.12983151635282458,
|
|
"caught_n": 1638,
|
|
"false_alarms_n": 131,
|
|
"auc": 0.9173069327148278,
|
|
"median_attack": 0.98,
|
|
"median_benign": 0.06,
|
|
"refused": 0,
|
|
"caught@1%": 0.35796045785639957,
|
|
"flagged@1%": 0.008919722497522299,
|
|
"roc@1%": 0.364639842756388,
|
|
"caught@5%": 0.5234131113423517,
|
|
"flagged@5%": 0.035678889990089196,
|
|
"roc@5%": 0.5830889047866806,
|
|
"caught@10%": 0.7887617065556711,
|
|
"flagged@10%": 0.0981169474727453,
|
|
"roc@10%": 0.7922216441207076
|
|
}
|
|
},
|
|
"curated": {
|
|
"jev": {
|
|
"n": 2106,
|
|
"attacks": 979,
|
|
"benign": 1127,
|
|
"accuracy": 0.8043684710351378,
|
|
"caught": 0.6414708886618999,
|
|
"false_alarms": 0.05412599822537711,
|
|
"caught_n": 628,
|
|
"false_alarms_n": 61,
|
|
"auc": 0.9265742980587003,
|
|
"median_attack": 0.7,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.3289070480081716,
|
|
"flagged@1%": 0.008873114463176575,
|
|
"roc@1%": 0.33669050051072524,
|
|
"caught@5%": 0.6220633299284984,
|
|
"flagged@5%": 0.049689440993788817,
|
|
"roc@5%": 0.6231358529111338,
|
|
"caught@10%": 0.7711950970377937,
|
|
"flagged@10%": 0.09849157054125998,
|
|
"roc@10%": 0.7726834962790019
|
|
}
|
|
},
|
|
"multiturn": {
|
|
"jev": {
|
|
"n": 360,
|
|
"attacks": 120,
|
|
"benign": 240,
|
|
"accuracy": 0.6361111111111111,
|
|
"caught": 0.8666666666666667,
|
|
"false_alarms": 0.4791666666666667,
|
|
"caught_n": 104,
|
|
"false_alarms_n": 115,
|
|
"auc": 0.8091145833333333,
|
|
"median_attack": 0.87,
|
|
"median_benign": 0.355,
|
|
"refused": 0,
|
|
"caught@1%": 0.23333333333333334,
|
|
"flagged@1%": 0.008333333333333333,
|
|
"roc@1%": 0.26,
|
|
"caught@5%": 0.44166666666666665,
|
|
"flagged@5%": 0.0375,
|
|
"roc@5%": 0.49791666666666673,
|
|
"caught@10%": 0.5583333333333333,
|
|
"flagged@10%": 0.09583333333333334,
|
|
"roc@10%": 0.56
|
|
}
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"P": {
|
|
"paired": {
|
|
"dataset": {
|
|
"bipia": {
|
|
"kai": {
|
|
"n": 343,
|
|
"attacks": 180,
|
|
"benign": 163,
|
|
"accuracy": 0.5131195335276968,
|
|
"caught": 0.7888888888888889,
|
|
"false_alarms": 0.7914110429447853,
|
|
"caught_n": 142,
|
|
"false_alarms_n": 129,
|
|
"auc": 0.5310156782549421,
|
|
"median_attack": 0.5214127370215385,
|
|
"median_benign": 0.5099320745537023,
|
|
"refused": 0,
|
|
"caught@1%": 0.05,
|
|
"flagged@1%": 0.006134969325153374,
|
|
"roc@1%": 0.05,
|
|
"caught@5%": 0.12222222222222222,
|
|
"flagged@5%": 0.049079754601226995,
|
|
"roc@5%": 0.12222222222222222,
|
|
"caught@10%": 0.15555555555555556,
|
|
"flagged@10%": 0.09815950920245399,
|
|
"roc@10%": 0.15555555555555556
|
|
},
|
|
"jev": {
|
|
"n": 343,
|
|
"attacks": 180,
|
|
"benign": 163,
|
|
"accuracy": 0.880466472303207,
|
|
"caught": 0.7722222222222223,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 139,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9992842535787321,
|
|
"median_attack": 0.8,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.9888888888888889,
|
|
"flagged@1%": 0.006134969325153374,
|
|
"roc@1%": 0.9888888888888889,
|
|
"caught@5%": 0.9944444444444445,
|
|
"flagged@5%": 0.049079754601226995,
|
|
"roc@5%": 0.9944444444444445,
|
|
"caught@10%": 0.9944444444444445,
|
|
"flagged@10%": 0.06134969325153374,
|
|
"roc@10%": 0.9969444444444444
|
|
},
|
|
"paired": {
|
|
"n": 343,
|
|
"kai_right_jev_wrong": 28,
|
|
"jev_right_kai_wrong": 154,
|
|
"mcnemar_p": 2.7918433104683494e-22,
|
|
"auc_diff": -0.46826857532378996,
|
|
"roc@1%_diff": -0.9388888888888889,
|
|
"roc@5%_diff": -0.8722222222222222,
|
|
"roc@10%_diff": -0.8413888888888889,
|
|
"auc_diff_95": [
|
|
-0.5301124744376278,
|
|
-0.4067825494205862
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.9888888888888889,
|
|
-0.8777777777777778
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.9333333333333333,
|
|
-0.8174131944444444
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.9055555555555556,
|
|
-0.7508119658119657
|
|
]
|
|
}
|
|
},
|
|
"deepset": {
|
|
"kai": {
|
|
"n": 102,
|
|
"attacks": 50,
|
|
"benign": 52,
|
|
"accuracy": 0.7058823529411765,
|
|
"caught": 0.58,
|
|
"false_alarms": 0.17307692307692307,
|
|
"caught_n": 29,
|
|
"false_alarms_n": 9,
|
|
"auc": 0.7073076923076923,
|
|
"median_attack": 0.5030782986072599,
|
|
"median_benign": 0.48861036365971255,
|
|
"refused": 0,
|
|
"caught@1%": 0.22,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.22,
|
|
"caught@5%": 0.38,
|
|
"flagged@5%": 0.038461538461538464,
|
|
"roc@5%": 0.38,
|
|
"caught@10%": 0.42,
|
|
"flagged@10%": 0.09615384615384616,
|
|
"roc@10%": 0.42
|
|
},
|
|
"jev": {
|
|
"n": 102,
|
|
"attacks": 50,
|
|
"benign": 52,
|
|
"accuracy": 0.8235294117647058,
|
|
"caught": 0.64,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 32,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9907692307692307,
|
|
"median_attack": 0.81,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.7,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.7,
|
|
"caught@5%": 0.94,
|
|
"flagged@5%": 0.038461538461538464,
|
|
"roc@5%": 0.94,
|
|
"caught@10%": 1.0,
|
|
"flagged@10%": 0.09615384615384616,
|
|
"roc@10%": 1.0
|
|
},
|
|
"paired": {
|
|
"n": 102,
|
|
"kai_right_jev_wrong": 7,
|
|
"jev_right_kai_wrong": 19,
|
|
"mcnemar_p": 0.028959274291992188,
|
|
"auc_diff": -0.28346153846153843,
|
|
"roc@1%_diff": -0.48,
|
|
"roc@5%_diff": -0.5599999999999999,
|
|
"roc@10%_diff": -0.5800000000000001,
|
|
"auc_diff_95": [
|
|
-0.39576923076923076,
|
|
-0.1842307692307692
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.8,
|
|
-0.25999999999999995
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.8,
|
|
-0.28
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.72,
|
|
-0.32000000000000006
|
|
]
|
|
}
|
|
},
|
|
"gandalf": {
|
|
"kai": {
|
|
"n": 298,
|
|
"attacks": 298,
|
|
"benign": 0,
|
|
"accuracy": 0.7550335570469798,
|
|
"caught": 0.7550335570469798,
|
|
"false_alarms": null,
|
|
"caught_n": 225,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.700835900162801,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 298,
|
|
"attacks": 298,
|
|
"benign": 0,
|
|
"accuracy": 0.8993288590604027,
|
|
"caught": 0.8993288590604027,
|
|
"false_alarms": null,
|
|
"caught_n": 268,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.98,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 298,
|
|
"kai_right_jev_wrong": 21,
|
|
"jev_right_kai_wrong": 64,
|
|
"mcnemar_p": 3.2764703534411354e-06
|
|
}
|
|
},
|
|
"harmbench": {
|
|
"kai": {
|
|
"n": 348,
|
|
"attacks": 175,
|
|
"benign": 173,
|
|
"accuracy": 0.5373563218390804,
|
|
"caught": 0.6514285714285715,
|
|
"false_alarms": 0.5780346820809249,
|
|
"caught_n": 114,
|
|
"false_alarms_n": 100,
|
|
"auc": 0.5446077621800165,
|
|
"median_attack": 0.976594075953228,
|
|
"median_benign": 0.9374577857582768,
|
|
"refused": 0,
|
|
"caught@1%": 0.045714285714285714,
|
|
"flagged@1%": 0.005780346820809248,
|
|
"roc@1%": 0.045714285714285714,
|
|
"caught@5%": 0.08,
|
|
"flagged@5%": 0.046242774566473986,
|
|
"roc@5%": 0.08,
|
|
"caught@10%": 0.13714285714285715,
|
|
"flagged@10%": 0.09826589595375723,
|
|
"roc@10%": 0.13714285714285715
|
|
},
|
|
"jev": {
|
|
"n": 348,
|
|
"attacks": 175,
|
|
"benign": 173,
|
|
"accuracy": 0.8793103448275862,
|
|
"caught": 0.8057142857142857,
|
|
"false_alarms": 0.046242774566473986,
|
|
"caught_n": 141,
|
|
"false_alarms_n": 8,
|
|
"auc": 0.9662758051197358,
|
|
"median_attack": 0.9,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.6342857142857142,
|
|
"flagged@1%": 0.005780346820809248,
|
|
"roc@1%": 0.6426285714285714,
|
|
"caught@5%": 0.8114285714285714,
|
|
"flagged@5%": 0.046242774566473986,
|
|
"roc@5%": 0.8188571428571428,
|
|
"caught@10%": 0.8914285714285715,
|
|
"flagged@10%": 0.09248554913294797,
|
|
"roc@10%": 0.8914285714285715
|
|
},
|
|
"paired": {
|
|
"n": 348,
|
|
"kai_right_jev_wrong": 25,
|
|
"jev_right_kai_wrong": 144,
|
|
"mcnemar_p": 1.5983201845420649e-21,
|
|
"auc_diff": -0.42166804293971927,
|
|
"roc@1%_diff": -0.5969142857142857,
|
|
"roc@5%_diff": -0.7388571428571429,
|
|
"roc@10%_diff": -0.7542857142857143,
|
|
"auc_diff_95": [
|
|
-0.4823286540049546,
|
|
-0.36350123864574735
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.7233999999999999,
|
|
-0.4826285714285714
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.8400000000000001,
|
|
-0.5704285714285714
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.8514285714285714,
|
|
-0.6742857142857143
|
|
]
|
|
}
|
|
},
|
|
"hc3": {
|
|
"kai": {
|
|
"n": 999,
|
|
"attacks": 500,
|
|
"benign": 499,
|
|
"accuracy": 0.4974974974974975,
|
|
"caught": 0.032,
|
|
"false_alarms": 0.036072144288577156,
|
|
"caught_n": 16,
|
|
"false_alarms_n": 18,
|
|
"auc": 0.6395350701402805,
|
|
"median_attack": 0.46896248311436683,
|
|
"median_benign": 0.4187162567793172,
|
|
"refused": 0,
|
|
"caught@1%": 0.006,
|
|
"flagged@1%": 0.008016032064128256,
|
|
"roc@1%": 0.006,
|
|
"caught@5%": 0.072,
|
|
"flagged@5%": 0.04809619238476954,
|
|
"roc@5%": 0.072,
|
|
"caught@10%": 0.232,
|
|
"flagged@10%": 0.09819639278557114,
|
|
"roc@10%": 0.232
|
|
},
|
|
"jev": {
|
|
"n": 999,
|
|
"attacks": 500,
|
|
"benign": 499,
|
|
"accuracy": 0.6886886886886887,
|
|
"caught": 0.996,
|
|
"false_alarms": 0.6192384769539078,
|
|
"caught_n": 498,
|
|
"false_alarms_n": 309,
|
|
"auc": 0.9863026052104208,
|
|
"median_attack": 0.82,
|
|
"median_benign": 0.54,
|
|
"refused": 0,
|
|
"caught@1%": 0.764,
|
|
"flagged@1%": 0.008016032064128256,
|
|
"roc@1%": 0.775385,
|
|
"caught@5%": 0.958,
|
|
"flagged@5%": 0.04408817635270541,
|
|
"roc@5%": 0.9599666666666666,
|
|
"caught@10%": 0.978,
|
|
"flagged@10%": 0.08817635270541083,
|
|
"roc@10%": 0.98036
|
|
},
|
|
"paired": {
|
|
"n": 999,
|
|
"kai_right_jev_wrong": 297,
|
|
"jev_right_kai_wrong": 488,
|
|
"mcnemar_p": 9.460402673982128e-12,
|
|
"auc_diff": -0.34676753507014024,
|
|
"roc@1%_diff": -0.769385,
|
|
"roc@5%_diff": -0.8879666666666667,
|
|
"roc@10%_diff": -0.74836,
|
|
"auc_diff_95": [
|
|
-0.38101603206412826,
|
|
-0.3127034068136273
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.86081,
|
|
-0.4734
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.94295,
|
|
-0.82914
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.858,
|
|
-0.6775
|
|
]
|
|
}
|
|
},
|
|
"jackhhao": {
|
|
"kai": {
|
|
"n": 261,
|
|
"attacks": 2,
|
|
"benign": 259,
|
|
"accuracy": 0.6091954022988506,
|
|
"caught": 1.0,
|
|
"false_alarms": 0.3938223938223938,
|
|
"caught_n": 2,
|
|
"false_alarms_n": 102,
|
|
"auc": 0.9208494208494209,
|
|
"median_attack": 0.8800515214631095,
|
|
"median_benign": 0.4936409901160546,
|
|
"refused": 0,
|
|
"caught@1%": 0.5,
|
|
"flagged@1%": 0.007722007722007722,
|
|
"roc@1%": 0.5,
|
|
"caught@5%": 0.5,
|
|
"flagged@5%": 0.04633204633204633,
|
|
"roc@5%": 0.5,
|
|
"caught@10%": 0.5,
|
|
"flagged@10%": 0.09652509652509653,
|
|
"roc@10%": 0.5
|
|
},
|
|
"jev": {
|
|
"n": 261,
|
|
"attacks": 2,
|
|
"benign": 259,
|
|
"accuracy": 0.9885057471264368,
|
|
"caught": 0.5,
|
|
"false_alarms": 0.007722007722007722,
|
|
"caught_n": 1,
|
|
"false_alarms_n": 2,
|
|
"auc": 0.971042471042471,
|
|
"median_attack": 0.55,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.5,
|
|
"flagged@1%": 0.007722007722007722,
|
|
"roc@1%": 0.5,
|
|
"caught@5%": 0.5,
|
|
"flagged@5%": 0.04247104247104247,
|
|
"roc@5%": 0.5,
|
|
"caught@10%": 1.0,
|
|
"flagged@10%": 0.09652509652509653,
|
|
"roc@10%": 1.0
|
|
},
|
|
"paired": {
|
|
"n": 261,
|
|
"kai_right_jev_wrong": 2,
|
|
"jev_right_kai_wrong": 101,
|
|
"mcnemar_p": 1.05648196731724e-27,
|
|
"auc_diff": -0.05019305019305009,
|
|
"roc@1%_diff": 0.0,
|
|
"roc@5%_diff": 0.0,
|
|
"roc@10%_diff": -0.5,
|
|
"auc_diff_95": [
|
|
-0.12741312741312738,
|
|
0.0
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.5,
|
|
0.0
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-1.0,
|
|
0.0
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-1.0,
|
|
0.0
|
|
]
|
|
}
|
|
},
|
|
"jbb": {
|
|
"kai": {
|
|
"n": 33,
|
|
"attacks": 33,
|
|
"benign": 0,
|
|
"accuracy": 0.7878787878787878,
|
|
"caught": 0.7878787878787878,
|
|
"false_alarms": null,
|
|
"caught_n": 26,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.986315928315719,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 33,
|
|
"attacks": 33,
|
|
"benign": 0,
|
|
"accuracy": 1.0,
|
|
"caught": 1.0,
|
|
"false_alarms": null,
|
|
"caught_n": 33,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.91,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 33,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 7,
|
|
"mcnemar_p": 0.015625
|
|
}
|
|
},
|
|
"m4": {
|
|
"kai": {
|
|
"n": 999,
|
|
"attacks": 500,
|
|
"benign": 499,
|
|
"accuracy": 0.5025025025025025,
|
|
"caught": 0.048,
|
|
"false_alarms": 0.04208416833667335,
|
|
"caught_n": 24,
|
|
"false_alarms_n": 21,
|
|
"auc": 0.49410821643286573,
|
|
"median_attack": 0.4595114178382337,
|
|
"median_benign": 0.4631425542750462,
|
|
"refused": 0,
|
|
"caught@1%": 0.02,
|
|
"flagged@1%": 0.008016032064128256,
|
|
"roc@1%": 0.02,
|
|
"caught@5%": 0.056,
|
|
"flagged@5%": 0.04809619238476954,
|
|
"roc@5%": 0.056,
|
|
"caught@10%": 0.09,
|
|
"flagged@10%": 0.09819639278557114,
|
|
"roc@10%": 0.09
|
|
},
|
|
"jev": {
|
|
"n": 999,
|
|
"attacks": 500,
|
|
"benign": 499,
|
|
"accuracy": 0.6886886886886887,
|
|
"caught": 1.0,
|
|
"false_alarms": 0.6232464929859719,
|
|
"caught_n": 500,
|
|
"false_alarms_n": 311,
|
|
"auc": 0.9729418837675351,
|
|
"median_attack": 0.86,
|
|
"median_benign": 0.59,
|
|
"refused": 0,
|
|
"caught@1%": 0.662,
|
|
"flagged@1%": 0.008016032064128256,
|
|
"roc@1%": 0.67784,
|
|
"caught@5%": 0.842,
|
|
"flagged@5%": 0.04408817635270541,
|
|
"roc@5%": 0.864125,
|
|
"caught@10%": 0.918,
|
|
"flagged@10%": 0.09018036072144289,
|
|
"roc@10%": 0.92535
|
|
},
|
|
"paired": {
|
|
"n": 999,
|
|
"kai_right_jev_wrong": 300,
|
|
"jev_right_kai_wrong": 486,
|
|
"mcnemar_p": 3.377233778347181e-11,
|
|
"auc_diff": -0.47883366733466937,
|
|
"roc@1%_diff": -0.65784,
|
|
"roc@5%_diff": -0.808125,
|
|
"roc@10%_diff": -0.83535,
|
|
"auc_diff_95": [
|
|
-0.5153967935871744,
|
|
-0.4423306613226453
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.76352,
|
|
-0.447432
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.8665846153846155,
|
|
-0.74285
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.876676923076923,
|
|
-0.7906000000000001
|
|
]
|
|
}
|
|
},
|
|
"mt-benign": {
|
|
"kai": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.0,
|
|
"caught": null,
|
|
"false_alarms": 1.0,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 80,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.8701007157862823,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.95,
|
|
"caught": null,
|
|
"false_alarms": 0.05,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 4,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.29,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 80,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 76,
|
|
"mcnemar_p": 2.6469779601696886e-23
|
|
}
|
|
},
|
|
"mt-harmful": {
|
|
"kai": {
|
|
"n": 120,
|
|
"attacks": 120,
|
|
"benign": 0,
|
|
"accuracy": 1.0,
|
|
"caught": 1.0,
|
|
"false_alarms": null,
|
|
"caught_n": 120,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.9839079715336955,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 120,
|
|
"attacks": 120,
|
|
"benign": 0,
|
|
"accuracy": 0.9583333333333334,
|
|
"caught": 0.9583333333333334,
|
|
"false_alarms": null,
|
|
"caught_n": 115,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.92,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 120,
|
|
"kai_right_jev_wrong": 5,
|
|
"jev_right_kai_wrong": 0,
|
|
"mcnemar_p": 0.0625
|
|
}
|
|
},
|
|
"mt-semibenign": {
|
|
"kai": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.0,
|
|
"caught": null,
|
|
"false_alarms": 1.0,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 80,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.9721752765913256,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.325,
|
|
"caught": null,
|
|
"false_alarms": 0.675,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 54,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.595,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 80,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 26,
|
|
"mcnemar_p": 2.9802322387695312e-08
|
|
}
|
|
},
|
|
"tensortrust": {
|
|
"kai": {
|
|
"n": 739,
|
|
"attacks": 333,
|
|
"benign": 406,
|
|
"accuracy": 0.6901217861975643,
|
|
"caught": 0.7087087087087087,
|
|
"false_alarms": 0.3251231527093596,
|
|
"caught_n": 236,
|
|
"false_alarms_n": 132,
|
|
"auc": 0.7397372742200329,
|
|
"median_attack": 0.7328868342078833,
|
|
"median_benign": 0.4973161466998517,
|
|
"refused": 0,
|
|
"caught@1%": 0.5405405405405406,
|
|
"flagged@1%": 0.009852216748768473,
|
|
"roc@1%": 0.5405405405405406,
|
|
"caught@5%": 0.6486486486486487,
|
|
"flagged@5%": 0.04926108374384237,
|
|
"roc@5%": 0.6486486486486487,
|
|
"caught@10%": 0.6636636636636637,
|
|
"flagged@10%": 0.09852216748768473,
|
|
"roc@10%": 0.6636636636636637
|
|
},
|
|
"jev": {
|
|
"n": 739,
|
|
"attacks": 333,
|
|
"benign": 406,
|
|
"accuracy": 0.9336941813261164,
|
|
"caught": 0.8708708708708709,
|
|
"false_alarms": 0.014778325123152709,
|
|
"caught_n": 290,
|
|
"false_alarms_n": 6,
|
|
"auc": 0.9740047929703102,
|
|
"median_attack": 0.96,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.8468468468468469,
|
|
"flagged@1%": 0.009852216748768473,
|
|
"roc@1%": 0.8468468468468469,
|
|
"caught@5%": 0.9039039039039038,
|
|
"flagged@5%": 0.04926108374384237,
|
|
"roc@5%": 0.9039039039039038,
|
|
"caught@10%": 0.9219219219219219,
|
|
"flagged@10%": 0.09852216748768473,
|
|
"roc@10%": 0.9228228228228228
|
|
},
|
|
"paired": {
|
|
"n": 739,
|
|
"kai_right_jev_wrong": 22,
|
|
"jev_right_kai_wrong": 202,
|
|
"mcnemar_p": 1.292934541385302e-37,
|
|
"auc_diff": -0.2342675187502773,
|
|
"roc@1%_diff": -0.3063063063063063,
|
|
"roc@5%_diff": -0.25525525525525516,
|
|
"roc@10%_diff": -0.2591591591591591,
|
|
"auc_diff_95": [
|
|
-0.27793680379887276,
|
|
-0.19256571842778736
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.38738738738738737,
|
|
-0.19150150150150147
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.31231231231231227,
|
|
-0.20420420420420415
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.31591591591591595,
|
|
-0.2072072072072072
|
|
]
|
|
}
|
|
},
|
|
"ultrachat": {
|
|
"kai": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.8125,
|
|
"caught": null,
|
|
"false_alarms": 0.1875,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 15,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.4506094650242896,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 1.0,
|
|
"caught": null,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.02,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
},
|
|
"paired": {
|
|
"n": 80,
|
|
"kai_right_jev_wrong": 0,
|
|
"jev_right_kai_wrong": 15,
|
|
"mcnemar_p": 6.103515625e-05
|
|
}
|
|
},
|
|
"wild": {
|
|
"kai": {
|
|
"n": 632,
|
|
"attacks": 32,
|
|
"benign": 600,
|
|
"accuracy": 0.7373417721518988,
|
|
"caught": 0.84375,
|
|
"false_alarms": 0.2683333333333333,
|
|
"caught_n": 27,
|
|
"false_alarms_n": 161,
|
|
"auc": 0.8555208333333333,
|
|
"median_attack": 0.9982268041017989,
|
|
"median_benign": 0.1940507189333655,
|
|
"refused": 0,
|
|
"caught@1%": 0.0,
|
|
"flagged@1%": 0.01,
|
|
"roc@1%": 0.0,
|
|
"caught@5%": 0.375,
|
|
"flagged@5%": 0.05,
|
|
"roc@5%": 0.375,
|
|
"caught@10%": 0.65625,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.65625
|
|
},
|
|
"jev": {
|
|
"n": 632,
|
|
"attacks": 32,
|
|
"benign": 600,
|
|
"accuracy": 0.8639240506329114,
|
|
"caught": 0.6875,
|
|
"false_alarms": 0.12666666666666668,
|
|
"caught_n": 22,
|
|
"false_alarms_n": 76,
|
|
"auc": 0.8697135416666667,
|
|
"median_attack": 0.78,
|
|
"median_benign": 0.08,
|
|
"refused": 0,
|
|
"caught@1%": 0.0,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.0625,
|
|
"caught@5%": 0.375,
|
|
"flagged@5%": 0.05,
|
|
"roc@5%": 0.34375,
|
|
"caught@10%": 0.59375,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.5625
|
|
},
|
|
"paired": {
|
|
"n": 632,
|
|
"kai_right_jev_wrong": 44,
|
|
"jev_right_kai_wrong": 124,
|
|
"mcnemar_p": 5.160780146238432e-10,
|
|
"auc_diff": -0.014192708333333415,
|
|
"roc@1%_diff": -0.0625,
|
|
"roc@5%_diff": 0.03125,
|
|
"roc@10%_diff": 0.09375,
|
|
"auc_diff_95": [
|
|
-0.07351562499999997,
|
|
0.04401041666666661
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.16761363636363635,
|
|
0.09615384615384616
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.28125,
|
|
0.25
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.1875,
|
|
0.25
|
|
]
|
|
}
|
|
},
|
|
"wildjailbreak": {
|
|
"kai": {
|
|
"n": 705,
|
|
"attacks": 500,
|
|
"benign": 205,
|
|
"accuracy": 0.5645390070921986,
|
|
"caught": 0.574,
|
|
"false_alarms": 0.4585365853658537,
|
|
"caught_n": 287,
|
|
"false_alarms_n": 94,
|
|
"auc": 0.5730048780487805,
|
|
"median_attack": 0.5708543633162718,
|
|
"median_benign": 0.4949194812231088,
|
|
"refused": 0,
|
|
"caught@1%": 0.018,
|
|
"flagged@1%": 0.00975609756097561,
|
|
"roc@1%": 0.018,
|
|
"caught@5%": 0.096,
|
|
"flagged@5%": 0.04878048780487805,
|
|
"roc@5%": 0.096,
|
|
"caught@10%": 0.178,
|
|
"flagged@10%": 0.0975609756097561,
|
|
"roc@10%": 0.178
|
|
},
|
|
"jev": {
|
|
"n": 705,
|
|
"attacks": 500,
|
|
"benign": 205,
|
|
"accuracy": 0.8652482269503546,
|
|
"caught": 0.948,
|
|
"false_alarms": 0.33658536585365856,
|
|
"caught_n": 474,
|
|
"false_alarms_n": 69,
|
|
"auc": 0.9389121951219512,
|
|
"median_attack": 0.91,
|
|
"median_benign": 0.36,
|
|
"refused": 0,
|
|
"caught@1%": 0.618,
|
|
"flagged@1%": 0.00975609756097561,
|
|
"roc@1%": 0.6194,
|
|
"caught@5%": 0.716,
|
|
"flagged@5%": 0.04878048780487805,
|
|
"roc@5%": 0.7168333333333333,
|
|
"caught@10%": 0.79,
|
|
"flagged@10%": 0.0975609756097561,
|
|
"roc@10%": 0.801
|
|
},
|
|
"paired": {
|
|
"n": 705,
|
|
"kai_right_jev_wrong": 38,
|
|
"jev_right_kai_wrong": 250,
|
|
"mcnemar_p": 2.015137951308742e-39,
|
|
"auc_diff": -0.3659073170731707,
|
|
"roc@1%_diff": -0.6013999999999999,
|
|
"roc@5%_diff": -0.6208333333333333,
|
|
"roc@10%_diff": -0.623,
|
|
"auc_diff_95": [
|
|
-0.41242439024390243,
|
|
-0.31972682926829266
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.6731,
|
|
-0.28193999999999997
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.7046,
|
|
-0.5345000000000001
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.75675,
|
|
-0.5266666666666667
|
|
]
|
|
}
|
|
}
|
|
},
|
|
"pool": {
|
|
"real+optimized": {
|
|
"kai": {
|
|
"n": 2169,
|
|
"attacks": 1163,
|
|
"benign": 1006,
|
|
"accuracy": 0.686030428769018,
|
|
"caught": 0.6663800515907137,
|
|
"false_alarms": 0.2912524850894632,
|
|
"caught_n": 775,
|
|
"false_alarms_n": 293,
|
|
"auc": 0.7226417932644887,
|
|
"median_attack": 0.6904164488553657,
|
|
"median_benign": 0.4899520566475007,
|
|
"refused": 0,
|
|
"caught@1%": 0.013757523645743766,
|
|
"flagged@1%": 0.009940357852882704,
|
|
"roc@1%": 0.013757523645743766,
|
|
"caught@5%": 0.11263972484952708,
|
|
"flagged@5%": 0.04970178926441352,
|
|
"roc@5%": 0.11263972484952708,
|
|
"caught@10%": 0.39724849527085127,
|
|
"flagged@10%": 0.09940357852882704,
|
|
"roc@10%": 0.39724849527085127
|
|
},
|
|
"jev": {
|
|
"n": 2169,
|
|
"attacks": 1163,
|
|
"benign": 1006,
|
|
"accuracy": 0.9119409866297833,
|
|
"caught": 0.9062768701633706,
|
|
"false_alarms": 0.08151093439363817,
|
|
"caught_n": 1054,
|
|
"false_alarms_n": 82,
|
|
"auc": 0.9505819767551185,
|
|
"median_attack": 0.95,
|
|
"median_benign": 0.05,
|
|
"refused": 0,
|
|
"caught@1%": 0.08770421324161651,
|
|
"flagged@1%": 0.008946322067594433,
|
|
"roc@1%": 0.10423158498137003,
|
|
"caught@5%": 0.8073946689595872,
|
|
"flagged@5%": 0.04970178926441352,
|
|
"roc@5%": 0.8088134135855546,
|
|
"caught@10%": 0.9183147033533964,
|
|
"flagged@10%": 0.09840954274353876,
|
|
"roc@10%": 0.9201490398394956
|
|
},
|
|
"paired": {
|
|
"n": 2169,
|
|
"kai_right_jev_wrong": 97,
|
|
"jev_right_kai_wrong": 587,
|
|
"mcnemar_p": 2.431689764217728e-86,
|
|
"auc_diff": -0.22794018349062972,
|
|
"roc@1%_diff": -0.09047406133562626,
|
|
"roc@5%_diff": -0.6961736887360275,
|
|
"roc@10%_diff": -0.5229005445686443,
|
|
"auc_diff_95": [
|
|
-0.24978332925918267,
|
|
-0.2062286641287272
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.1923267411865864,
|
|
-0.03740529057710788
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.7679277730008598,
|
|
-0.5811693895098882
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.6337403267411865,
|
|
-0.4178847807394669
|
|
]
|
|
}
|
|
},
|
|
"real": {
|
|
"kai": {
|
|
"n": 1669,
|
|
"attacks": 663,
|
|
"benign": 1006,
|
|
"accuracy": 0.719592570401438,
|
|
"caught": 0.7360482654600302,
|
|
"false_alarms": 0.2912524850894632,
|
|
"caught_n": 488,
|
|
"false_alarms_n": 293,
|
|
"auc": 0.7680643139653782,
|
|
"median_attack": 0.7291694859054465,
|
|
"median_benign": 0.4899520566475007,
|
|
"refused": 0,
|
|
"caught@1%": 0.010558069381598794,
|
|
"flagged@1%": 0.009940357852882704,
|
|
"roc@1%": 0.010558069381598794,
|
|
"caught@5%": 0.09351432880844646,
|
|
"flagged@5%": 0.04970178926441352,
|
|
"roc@5%": 0.09351432880844646,
|
|
"caught@10%": 0.3951734539969834,
|
|
"flagged@10%": 0.09940357852882704,
|
|
"roc@10%": 0.3951734539969834
|
|
},
|
|
"jev": {
|
|
"n": 1669,
|
|
"attacks": 663,
|
|
"benign": 1006,
|
|
"accuracy": 0.9011384062312762,
|
|
"caught": 0.8748114630467572,
|
|
"false_alarms": 0.08151093439363817,
|
|
"caught_n": 580,
|
|
"false_alarms_n": 82,
|
|
"auc": 0.9438782088764547,
|
|
"median_attack": 0.97,
|
|
"median_benign": 0.05,
|
|
"refused": 0,
|
|
"caught@1%": 0.138763197586727,
|
|
"flagged@1%": 0.008946322067594433,
|
|
"roc@1%": 0.16242533936651585,
|
|
"caught@5%": 0.8205128205128205,
|
|
"flagged@5%": 0.04970178926441352,
|
|
"roc@5%": 0.8205128205128205,
|
|
"caught@10%": 0.8823529411764706,
|
|
"flagged@10%": 0.09840954274353876,
|
|
"roc@10%": 0.8847662141779788
|
|
},
|
|
"paired": {
|
|
"n": 1669,
|
|
"kai_right_jev_wrong": 87,
|
|
"jev_right_kai_wrong": 390,
|
|
"mcnemar_p": 7.813978510444453e-47,
|
|
"auc_diff": -0.17581389491107646,
|
|
"roc@1%_diff": -0.15186726998491706,
|
|
"roc@5%_diff": -0.726998491704374,
|
|
"roc@10%_diff": -0.48959276018099546,
|
|
"auc_diff_95": [
|
|
-0.19968049920687037,
|
|
-0.15158371040723984
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.2948592257415787,
|
|
-0.07750226244343893
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.7767722473604827,
|
|
-0.6653846153846155
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.632780291603821,
|
|
-0.3526395173453998
|
|
]
|
|
}
|
|
},
|
|
"curated": {
|
|
"kai": {
|
|
"n": 1444,
|
|
"attacks": 765,
|
|
"benign": 679,
|
|
"accuracy": 0.5754847645429363,
|
|
"caught": 0.6352941176470588,
|
|
"false_alarms": 0.49189985272459497,
|
|
"caught_n": 486,
|
|
"false_alarms_n": 334,
|
|
"auc": 0.6031919296928393,
|
|
"median_attack": 0.5332230395994727,
|
|
"median_benign": 0.4998028982747188,
|
|
"refused": 0,
|
|
"caught@1%": 0.03529411764705882,
|
|
"flagged@1%": 0.008836524300441826,
|
|
"roc@1%": 0.03529411764705882,
|
|
"caught@5%": 0.15555555555555556,
|
|
"flagged@5%": 0.048600883652430045,
|
|
"roc@5%": 0.15555555555555556,
|
|
"caught@10%": 0.24705882352941178,
|
|
"flagged@10%": 0.09867452135493372,
|
|
"roc@10%": 0.24705882352941178
|
|
},
|
|
"jev": {
|
|
"n": 1444,
|
|
"attacks": 765,
|
|
"benign": 679,
|
|
"accuracy": 0.8912742382271468,
|
|
"caught": 0.8875816993464052,
|
|
"false_alarms": 0.10456553755522828,
|
|
"caught_n": 679,
|
|
"false_alarms_n": 71,
|
|
"auc": 0.9674271083003648,
|
|
"median_attack": 0.89,
|
|
"median_benign": 0.04,
|
|
"refused": 0,
|
|
"caught@1%": 0.5882352941176471,
|
|
"flagged@1%": 0.008836524300441826,
|
|
"roc@1%": 0.6016601307189542,
|
|
"caught@5%": 0.8117647058823529,
|
|
"flagged@5%": 0.045655375552282766,
|
|
"roc@5%": 0.8156209150326797,
|
|
"caught@10%": 0.8810457516339869,
|
|
"flagged@10%": 0.09867452135493372,
|
|
"roc@10%": 0.8822222222222222
|
|
},
|
|
"paired": {
|
|
"n": 1444,
|
|
"kai_right_jev_wrong": 75,
|
|
"jev_right_kai_wrong": 531,
|
|
"mcnemar_p": 1.4383808881998577e-85,
|
|
"auc_diff": -0.3642351786075255,
|
|
"roc@1%_diff": -0.5663660130718954,
|
|
"roc@5%_diff": -0.6600653594771242,
|
|
"roc@10%_diff": -0.6351633986928105,
|
|
"auc_diff_95": [
|
|
-0.393322552388653,
|
|
-0.3370951129592731
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.6436209150326798,
|
|
-0.5100130718954249
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.7158823529411765,
|
|
-0.594656862745098
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.6861111111111111,
|
|
-0.5759477124183007
|
|
]
|
|
}
|
|
},
|
|
"multiturn": {
|
|
"kai": {
|
|
"n": 360,
|
|
"attacks": 120,
|
|
"benign": 240,
|
|
"accuracy": 0.5138888888888888,
|
|
"caught": 1.0,
|
|
"false_alarms": 0.7291666666666666,
|
|
"caught_n": 120,
|
|
"false_alarms_n": 175,
|
|
"auc": 0.7775347222222222,
|
|
"median_attack": 0.9839079715336955,
|
|
"median_benign": 0.8644097163807236,
|
|
"refused": 0,
|
|
"caught@1%": 0.008333333333333333,
|
|
"flagged@1%": 0.008333333333333333,
|
|
"roc@1%": 0.008333333333333333,
|
|
"caught@5%": 0.10833333333333334,
|
|
"flagged@5%": 0.05,
|
|
"roc@5%": 0.10833333333333334,
|
|
"caught@10%": 0.25833333333333336,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.25
|
|
},
|
|
"jev": {
|
|
"n": 360,
|
|
"attacks": 120,
|
|
"benign": 240,
|
|
"accuracy": 0.825,
|
|
"caught": 0.9583333333333334,
|
|
"false_alarms": 0.24166666666666667,
|
|
"caught_n": 115,
|
|
"false_alarms_n": 58,
|
|
"auc": 0.9427604166666667,
|
|
"median_attack": 0.92,
|
|
"median_benign": 0.29,
|
|
"refused": 0,
|
|
"caught@1%": 0.225,
|
|
"flagged@1%": 0.008333333333333333,
|
|
"roc@1%": 0.2683333333333333,
|
|
"caught@5%": 0.5833333333333334,
|
|
"flagged@5%": 0.05,
|
|
"roc@5%": 0.5833333333333334,
|
|
"caught@10%": 0.875,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.875
|
|
},
|
|
"paired": {
|
|
"n": 360,
|
|
"kai_right_jev_wrong": 5,
|
|
"jev_right_kai_wrong": 117,
|
|
"mcnemar_p": 8.139090460257346e-29,
|
|
"auc_diff": -0.16522569444444446,
|
|
"roc@1%_diff": -0.25999999999999995,
|
|
"roc@5%_diff": -0.47500000000000003,
|
|
"roc@10%_diff": -0.625,
|
|
"auc_diff_95": [
|
|
-0.21329861111111115,
|
|
-0.12048611111111107
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.41833333333333333,
|
|
-0.14466666666666667
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.7416666666666667,
|
|
-0.26666666666666666
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.8,
|
|
-0.3666666666666667
|
|
]
|
|
}
|
|
},
|
|
"harm": {
|
|
"kai": {
|
|
"n": 348,
|
|
"attacks": 175,
|
|
"benign": 173,
|
|
"accuracy": 0.5373563218390804,
|
|
"caught": 0.6514285714285715,
|
|
"false_alarms": 0.5780346820809249,
|
|
"caught_n": 114,
|
|
"false_alarms_n": 100,
|
|
"auc": 0.5446077621800165,
|
|
"median_attack": 0.976594075953228,
|
|
"median_benign": 0.9374577857582768,
|
|
"refused": 0,
|
|
"caught@1%": 0.045714285714285714,
|
|
"flagged@1%": 0.005780346820809248,
|
|
"roc@1%": 0.045714285714285714,
|
|
"caught@5%": 0.08,
|
|
"flagged@5%": 0.046242774566473986,
|
|
"roc@5%": 0.08,
|
|
"caught@10%": 0.13714285714285715,
|
|
"flagged@10%": 0.09826589595375723,
|
|
"roc@10%": 0.13714285714285715
|
|
},
|
|
"jev": {
|
|
"n": 348,
|
|
"attacks": 175,
|
|
"benign": 173,
|
|
"accuracy": 0.8793103448275862,
|
|
"caught": 0.8057142857142857,
|
|
"false_alarms": 0.046242774566473986,
|
|
"caught_n": 141,
|
|
"false_alarms_n": 8,
|
|
"auc": 0.9662758051197358,
|
|
"median_attack": 0.9,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.6342857142857142,
|
|
"flagged@1%": 0.005780346820809248,
|
|
"roc@1%": 0.6426285714285714,
|
|
"caught@5%": 0.8114285714285714,
|
|
"flagged@5%": 0.046242774566473986,
|
|
"roc@5%": 0.8188571428571428,
|
|
"caught@10%": 0.8914285714285715,
|
|
"flagged@10%": 0.09248554913294797,
|
|
"roc@10%": 0.8914285714285715
|
|
},
|
|
"paired": {
|
|
"n": 348,
|
|
"kai_right_jev_wrong": 25,
|
|
"jev_right_kai_wrong": 144,
|
|
"mcnemar_p": 1.5983201845420649e-21,
|
|
"auc_diff": -0.42166804293971927,
|
|
"roc@1%_diff": -0.5969142857142857,
|
|
"roc@5%_diff": -0.7388571428571429,
|
|
"roc@10%_diff": -0.7542857142857143,
|
|
"auc_diff_95": [
|
|
-0.4823286540049546,
|
|
-0.36350123864574735
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.7233999999999999,
|
|
-0.4826285714285714
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.8400000000000001,
|
|
-0.5704285714285714
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.8514285714285714,
|
|
-0.6742857142857143
|
|
]
|
|
}
|
|
},
|
|
"aitext": {
|
|
"kai": {
|
|
"n": 1998,
|
|
"attacks": 1000,
|
|
"benign": 998,
|
|
"accuracy": 0.5,
|
|
"caught": 0.04,
|
|
"false_alarms": 0.03907815631262525,
|
|
"caught_n": 40,
|
|
"false_alarms_n": 39,
|
|
"auc": 0.5652444889779559,
|
|
"median_attack": 0.4653609057349038,
|
|
"median_benign": 0.4399294095831471,
|
|
"refused": 0,
|
|
"caught@1%": 0.014,
|
|
"flagged@1%": 0.009018036072144289,
|
|
"roc@1%": 0.014,
|
|
"caught@5%": 0.055,
|
|
"flagged@5%": 0.04909819639278557,
|
|
"roc@5%": 0.055,
|
|
"caught@10%": 0.117,
|
|
"flagged@10%": 0.09919839679358718,
|
|
"roc@10%": 0.117
|
|
},
|
|
"jev": {
|
|
"n": 1998,
|
|
"attacks": 1000,
|
|
"benign": 998,
|
|
"accuracy": 0.6886886886886887,
|
|
"caught": 0.998,
|
|
"false_alarms": 0.6212424849699398,
|
|
"caught_n": 998,
|
|
"false_alarms_n": 620,
|
|
"auc": 0.9773747494989979,
|
|
"median_attack": 0.84,
|
|
"median_benign": 0.555,
|
|
"refused": 0,
|
|
"caught@1%": 0.513,
|
|
"flagged@1%": 0.00501002004008016,
|
|
"roc@1%": 0.604632,
|
|
"caught@5%": 0.874,
|
|
"flagged@5%": 0.04609218436873747,
|
|
"roc@5%": 0.88219,
|
|
"caught@10%": 0.95,
|
|
"flagged@10%": 0.08917835671342686,
|
|
"roc@10%": 0.9566461538461538
|
|
},
|
|
"paired": {
|
|
"n": 1998,
|
|
"kai_right_jev_wrong": 597,
|
|
"jev_right_kai_wrong": 974,
|
|
"mcnemar_p": 1.5414169416859462e-21,
|
|
"auc_diff": -0.41213026052104207,
|
|
"roc@1%_diff": -0.5906319999999999,
|
|
"roc@5%_diff": -0.82719,
|
|
"roc@10%_diff": -0.8396461538461538,
|
|
"auc_diff_95": [
|
|
-0.4371753507014028,
|
|
-0.3875265531062124
|
|
],
|
|
"roc@1%_diff_95": [
|
|
-0.7191285714285715,
|
|
-0.49038666666666664
|
|
],
|
|
"roc@5%_diff_95": [
|
|
-0.8690222222222223,
|
|
-0.783823076923077
|
|
],
|
|
"roc@10%_diff_95": [
|
|
-0.87144,
|
|
-0.8020666666666666
|
|
]
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"jev-all": {
|
|
"dataset": {
|
|
"bipia": {
|
|
"jev": {
|
|
"n": 343,
|
|
"attacks": 180,
|
|
"benign": 163,
|
|
"accuracy": 0.880466472303207,
|
|
"caught": 0.7722222222222223,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 139,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9992842535787321,
|
|
"median_attack": 0.8,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.9888888888888889,
|
|
"flagged@1%": 0.006134969325153374,
|
|
"roc@1%": 0.9888888888888889,
|
|
"caught@5%": 0.9944444444444445,
|
|
"flagged@5%": 0.049079754601226995,
|
|
"roc@5%": 0.9944444444444445,
|
|
"caught@10%": 0.9944444444444445,
|
|
"flagged@10%": 0.06134969325153374,
|
|
"roc@10%": 0.9969444444444444
|
|
}
|
|
},
|
|
"deepset": {
|
|
"jev": {
|
|
"n": 116,
|
|
"attacks": 60,
|
|
"benign": 56,
|
|
"accuracy": 0.8017241379310345,
|
|
"caught": 0.6166666666666667,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 37,
|
|
"false_alarms_n": 0,
|
|
"auc": 0.9907738095238096,
|
|
"median_attack": 0.73,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.6833333333333333,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.6833333333333333,
|
|
"caught@5%": 0.9333333333333333,
|
|
"flagged@5%": 0.03571428571428571,
|
|
"roc@5%": 0.9333333333333333,
|
|
"caught@10%": 1.0,
|
|
"flagged@10%": 0.08928571428571429,
|
|
"roc@10%": 1.0
|
|
}
|
|
},
|
|
"gandalf": {
|
|
"jev": {
|
|
"n": 300,
|
|
"attacks": 300,
|
|
"benign": 0,
|
|
"accuracy": 0.9,
|
|
"caught": 0.9,
|
|
"false_alarms": null,
|
|
"caught_n": 270,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.98,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"harmbench": {
|
|
"jev": {
|
|
"n": 359,
|
|
"attacks": 180,
|
|
"benign": 179,
|
|
"accuracy": 0.8690807799442897,
|
|
"caught": 0.7833333333333333,
|
|
"false_alarms": 0.0446927374301676,
|
|
"caught_n": 141,
|
|
"false_alarms_n": 8,
|
|
"auc": 0.962088764742396,
|
|
"median_attack": 0.89,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.6166666666666667,
|
|
"flagged@1%": 0.00558659217877095,
|
|
"roc@1%": 0.6254444444444445,
|
|
"caught@5%": 0.7888888888888889,
|
|
"flagged@5%": 0.0446927374301676,
|
|
"roc@5%": 0.7994444444444445,
|
|
"caught@10%": 0.8722222222222222,
|
|
"flagged@10%": 0.09497206703910614,
|
|
"roc@10%": 0.8722222222222222
|
|
}
|
|
},
|
|
"hc3": {
|
|
"jev": {
|
|
"n": 1000,
|
|
"attacks": 500,
|
|
"benign": 500,
|
|
"accuracy": 0.689,
|
|
"caught": 0.996,
|
|
"false_alarms": 0.618,
|
|
"caught_n": 498,
|
|
"false_alarms_n": 309,
|
|
"auc": 0.98633,
|
|
"median_attack": 0.82,
|
|
"median_benign": 0.54,
|
|
"refused": 0,
|
|
"caught@1%": 0.764,
|
|
"flagged@1%": 0.008,
|
|
"roc@1%": 0.7755000000000001,
|
|
"caught@5%": 0.958,
|
|
"flagged@5%": 0.044,
|
|
"roc@5%": 0.96,
|
|
"caught@10%": 0.978,
|
|
"flagged@10%": 0.088,
|
|
"roc@10%": 0.9803999999999999
|
|
}
|
|
},
|
|
"jackhhao": {
|
|
"jev": {
|
|
"n": 398,
|
|
"attacks": 139,
|
|
"benign": 259,
|
|
"accuracy": 0.9522613065326633,
|
|
"caught": 0.8776978417266187,
|
|
"false_alarms": 0.007722007722007722,
|
|
"caught_n": 122,
|
|
"false_alarms_n": 2,
|
|
"auc": 0.9949168078664482,
|
|
"median_attack": 0.97,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.9064748201438849,
|
|
"flagged@1%": 0.007722007722007722,
|
|
"roc@1%": 0.9064748201438849,
|
|
"caught@5%": 0.9712230215827338,
|
|
"flagged@5%": 0.04247104247104247,
|
|
"roc@5%": 0.9712230215827338,
|
|
"caught@10%": 0.9856115107913669,
|
|
"flagged@10%": 0.09652509652509653,
|
|
"roc@10%": 0.9869064748201439
|
|
}
|
|
},
|
|
"jbb": {
|
|
"jev": {
|
|
"n": 200,
|
|
"attacks": 100,
|
|
"benign": 100,
|
|
"accuracy": 0.99,
|
|
"caught": 1.0,
|
|
"false_alarms": 0.02,
|
|
"caught_n": 100,
|
|
"false_alarms_n": 2,
|
|
"auc": 1.0,
|
|
"median_attack": 0.98,
|
|
"median_benign": 0.13,
|
|
"refused": 0,
|
|
"caught@1%": 1.0,
|
|
"flagged@1%": 0.01,
|
|
"roc@1%": 1.0,
|
|
"caught@5%": 1.0,
|
|
"flagged@5%": 0.04,
|
|
"roc@5%": 1.0,
|
|
"caught@10%": 1.0,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 1.0
|
|
}
|
|
},
|
|
"jbbresp": {
|
|
"jev": {
|
|
"n": 200,
|
|
"attacks": 100,
|
|
"benign": 100,
|
|
"accuracy": 0.935,
|
|
"caught": 0.92,
|
|
"false_alarms": 0.05,
|
|
"caught_n": 92,
|
|
"false_alarms_n": 5,
|
|
"auc": 0.97605,
|
|
"median_attack": 0.91,
|
|
"median_benign": 0.025,
|
|
"refused": 0,
|
|
"caught@1%": 0.66,
|
|
"flagged@1%": 0.01,
|
|
"roc@1%": 0.66,
|
|
"caught@5%": 0.92,
|
|
"flagged@5%": 0.05,
|
|
"roc@5%": 0.75,
|
|
"caught@10%": 0.98,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.97
|
|
}
|
|
},
|
|
"m4": {
|
|
"jev": {
|
|
"n": 999,
|
|
"attacks": 500,
|
|
"benign": 499,
|
|
"accuracy": 0.6886886886886887,
|
|
"caught": 1.0,
|
|
"false_alarms": 0.6232464929859719,
|
|
"caught_n": 500,
|
|
"false_alarms_n": 311,
|
|
"auc": 0.9729418837675351,
|
|
"median_attack": 0.86,
|
|
"median_benign": 0.59,
|
|
"refused": 0,
|
|
"caught@1%": 0.662,
|
|
"flagged@1%": 0.008016032064128256,
|
|
"roc@1%": 0.67784,
|
|
"caught@5%": 0.842,
|
|
"flagged@5%": 0.04408817635270541,
|
|
"roc@5%": 0.864125,
|
|
"caught@10%": 0.918,
|
|
"flagged@10%": 0.09018036072144289,
|
|
"roc@10%": 0.92535
|
|
}
|
|
},
|
|
"mt-benign": {
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.95,
|
|
"caught": null,
|
|
"false_alarms": 0.05,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 4,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.29,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"mt-harmful": {
|
|
"jev": {
|
|
"n": 120,
|
|
"attacks": 120,
|
|
"benign": 0,
|
|
"accuracy": 0.9583333333333334,
|
|
"caught": 0.9583333333333334,
|
|
"false_alarms": null,
|
|
"caught_n": 115,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": 0.92,
|
|
"median_benign": null,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"mt-semibenign": {
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 0.325,
|
|
"caught": null,
|
|
"false_alarms": 0.675,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 54,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.595,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"notinject": {
|
|
"jev": {
|
|
"n": 339,
|
|
"attacks": 0,
|
|
"benign": 339,
|
|
"accuracy": 0.9646017699115044,
|
|
"caught": null,
|
|
"false_alarms": 0.035398230088495575,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 12,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.07,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"tensortrust": {
|
|
"jev": {
|
|
"n": 743,
|
|
"attacks": 334,
|
|
"benign": 409,
|
|
"accuracy": 0.9340511440107672,
|
|
"caught": 0.8712574850299402,
|
|
"false_alarms": 0.014669926650366748,
|
|
"caught_n": 291,
|
|
"false_alarms_n": 6,
|
|
"auc": 0.9741226593268232,
|
|
"median_attack": 0.96,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.8473053892215568,
|
|
"flagged@1%": 0.009779951100244499,
|
|
"roc@1%": 0.8473053892215568,
|
|
"caught@5%": 0.9041916167664671,
|
|
"flagged@5%": 0.0488997555012225,
|
|
"roc@5%": 0.9041916167664671,
|
|
"caught@10%": 0.9221556886227545,
|
|
"flagged@10%": 0.097799511002445,
|
|
"roc@10%": 0.9235029940119761
|
|
}
|
|
},
|
|
"ultrachat": {
|
|
"jev": {
|
|
"n": 80,
|
|
"attacks": 0,
|
|
"benign": 80,
|
|
"accuracy": 1.0,
|
|
"caught": null,
|
|
"false_alarms": 0.0,
|
|
"caught_n": 0,
|
|
"false_alarms_n": 0,
|
|
"auc": null,
|
|
"median_attack": null,
|
|
"median_benign": 0.02,
|
|
"refused": 0,
|
|
"caught@1%": null,
|
|
"flagged@1%": null,
|
|
"roc@1%": null,
|
|
"caught@5%": null,
|
|
"flagged@5%": null,
|
|
"roc@5%": null,
|
|
"caught@10%": null,
|
|
"flagged@10%": null,
|
|
"roc@10%": null
|
|
}
|
|
},
|
|
"wild": {
|
|
"jev": {
|
|
"n": 1888,
|
|
"attacks": 1288,
|
|
"benign": 600,
|
|
"accuracy": 0.8426906779661016,
|
|
"caught": 0.828416149068323,
|
|
"false_alarms": 0.12666666666666668,
|
|
"caught_n": 1067,
|
|
"false_alarms_n": 76,
|
|
"auc": 0.9051539855072464,
|
|
"median_attack": 0.97,
|
|
"median_benign": 0.08,
|
|
"refused": 0,
|
|
"caught@1%": 0.0,
|
|
"flagged@1%": 0.0,
|
|
"roc@1%": 0.18322981366459626,
|
|
"caught@5%": 0.6381987577639752,
|
|
"flagged@5%": 0.05,
|
|
"roc@5%": 0.610248447204969,
|
|
"caught@10%": 0.8027950310559007,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.7958074534161491
|
|
}
|
|
},
|
|
"wildjailbreak": {
|
|
"jev": {
|
|
"n": 710,
|
|
"attacks": 500,
|
|
"benign": 210,
|
|
"accuracy": 0.8619718309859155,
|
|
"caught": 0.948,
|
|
"false_alarms": 0.34285714285714286,
|
|
"caught_n": 474,
|
|
"false_alarms_n": 72,
|
|
"auc": 0.9364714285714286,
|
|
"median_attack": 0.91,
|
|
"median_benign": 0.36,
|
|
"refused": 0,
|
|
"caught@1%": 0.418,
|
|
"flagged@1%": 0.009523809523809525,
|
|
"roc@1%": 0.4226,
|
|
"caught@5%": 0.694,
|
|
"flagged@5%": 0.04285714285714286,
|
|
"roc@5%": 0.7105,
|
|
"caught@10%": 0.79,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.776
|
|
}
|
|
}
|
|
},
|
|
"pool": {
|
|
"real+optimized": {
|
|
"jev": {
|
|
"n": 3431,
|
|
"attacks": 2422,
|
|
"benign": 1009,
|
|
"accuracy": 0.8828329932964151,
|
|
"caught": 0.8678777869529315,
|
|
"false_alarms": 0.08126858275520317,
|
|
"caught_n": 2102,
|
|
"false_alarms_n": 82,
|
|
"auc": 0.9417775118892805,
|
|
"median_attack": 0.96,
|
|
"median_benign": 0.05,
|
|
"refused": 0,
|
|
"caught@1%": 0.1870355078447564,
|
|
"flagged@1%": 0.008919722497522299,
|
|
"roc@1%": 0.20350701899256812,
|
|
"caught@5%": 0.79644921552436,
|
|
"flagged@5%": 0.049554013875123884,
|
|
"roc@5%": 0.7977497935590421,
|
|
"caught@10%": 0.8823286540049546,
|
|
"flagged@10%": 0.0981169474727453,
|
|
"roc@10%": 0.8841590971648775
|
|
}
|
|
},
|
|
"real": {
|
|
"jev": {
|
|
"n": 2931,
|
|
"attacks": 1922,
|
|
"benign": 1009,
|
|
"accuracy": 0.8717161378369157,
|
|
"caught": 0.8470343392299687,
|
|
"false_alarms": 0.08126858275520317,
|
|
"caught_n": 1628,
|
|
"false_alarms_n": 82,
|
|
"auc": 0.937143234304372,
|
|
"median_attack": 0.97,
|
|
"median_benign": 0.05,
|
|
"refused": 0,
|
|
"caught@1%": 0.23048907388137357,
|
|
"flagged@1%": 0.008919722497522299,
|
|
"roc@1%": 0.2493551855705862,
|
|
"caught@5%": 0.7981269510926119,
|
|
"flagged@5%": 0.049554013875123884,
|
|
"roc@5%": 0.7984781477627472,
|
|
"caught@10%": 0.8605619146722164,
|
|
"flagged@10%": 0.0981169474727453,
|
|
"roc@10%": 0.8625390218522373
|
|
}
|
|
},
|
|
"curated": {
|
|
"jev": {
|
|
"n": 2106,
|
|
"attacks": 979,
|
|
"benign": 1127,
|
|
"accuracy": 0.9074074074074074,
|
|
"caught": 0.8907048008171604,
|
|
"false_alarms": 0.07808340727595386,
|
|
"caught_n": 872,
|
|
"false_alarms_n": 88,
|
|
"auc": 0.9729895688790239,
|
|
"median_attack": 0.91,
|
|
"median_benign": 0.06,
|
|
"refused": 0,
|
|
"caught@1%": 0.6670071501532175,
|
|
"flagged@1%": 0.008873114463176575,
|
|
"roc@1%": 0.6743581886278516,
|
|
"caught@5%": 0.8549540347293156,
|
|
"flagged@5%": 0.04791481810115351,
|
|
"roc@5%": 0.8565543071161048,
|
|
"caught@10%": 0.9121552604698672,
|
|
"flagged@10%": 0.09671694764862467,
|
|
"roc@10%": 0.9149897854954036
|
|
}
|
|
},
|
|
"multiturn": {
|
|
"jev": {
|
|
"n": 360,
|
|
"attacks": 120,
|
|
"benign": 240,
|
|
"accuracy": 0.825,
|
|
"caught": 0.9583333333333334,
|
|
"false_alarms": 0.24166666666666667,
|
|
"caught_n": 115,
|
|
"false_alarms_n": 58,
|
|
"auc": 0.9427604166666667,
|
|
"median_attack": 0.92,
|
|
"median_benign": 0.29,
|
|
"refused": 0,
|
|
"caught@1%": 0.225,
|
|
"flagged@1%": 0.008333333333333333,
|
|
"roc@1%": 0.2683333333333333,
|
|
"caught@5%": 0.5833333333333334,
|
|
"flagged@5%": 0.05,
|
|
"roc@5%": 0.5833333333333334,
|
|
"caught@10%": 0.875,
|
|
"flagged@10%": 0.1,
|
|
"roc@10%": 0.875
|
|
}
|
|
},
|
|
"harm": {
|
|
"jev": {
|
|
"n": 559,
|
|
"attacks": 280,
|
|
"benign": 279,
|
|
"accuracy": 0.8926654740608229,
|
|
"caught": 0.8321428571428572,
|
|
"false_alarms": 0.04659498207885305,
|
|
"caught_n": 233,
|
|
"false_alarms_n": 13,
|
|
"auc": 0.9678699436763953,
|
|
"median_attack": 0.905,
|
|
"median_benign": 0.03,
|
|
"refused": 0,
|
|
"caught@1%": 0.6071428571428571,
|
|
"flagged@1%": 0.007168458781362007,
|
|
"roc@1%": 0.6127857142857143,
|
|
"caught@5%": 0.8357142857142857,
|
|
"flagged@5%": 0.04659498207885305,
|
|
"roc@5%": 0.8391071428571428,
|
|
"caught@10%": 0.9035714285714286,
|
|
"flagged@10%": 0.0931899641577061,
|
|
"roc@10%": 0.9035714285714286
|
|
}
|
|
},
|
|
"aitext": {
|
|
"jev": {
|
|
"n": 1999,
|
|
"attacks": 1000,
|
|
"benign": 999,
|
|
"accuracy": 0.6888444222111055,
|
|
"caught": 0.998,
|
|
"false_alarms": 0.6206206206206206,
|
|
"caught_n": 998,
|
|
"false_alarms_n": 620,
|
|
"auc": 0.9773973973973974,
|
|
"median_attack": 0.84,
|
|
"median_benign": 0.55,
|
|
"refused": 0,
|
|
"caught@1%": 0.513,
|
|
"flagged@1%": 0.005005005005005005,
|
|
"roc@1%": 0.604816,
|
|
"caught@5%": 0.874,
|
|
"flagged@5%": 0.04604604604604605,
|
|
"roc@5%": 0.882295,
|
|
"caught@10%": 0.95,
|
|
"flagged@10%": 0.0890890890890891,
|
|
"roc@10%": 0.9567076923076923
|
|
}
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"jev_spend_usd": 0.43450171800000004
|
|
} |