{
  "date": "2026-09-11",
  "organization": "CodeIntegrity",
  "hardware": "Apple M5 Pro, 24 GB RAM; PyTorch MPS; float32",
  "protocol": "PROTOCOL.md",
  "models": [
    {
      "id": "codeintegrity-ai/promptguard",
      "name": "CodeIntegrity PromptGuard",
      "revision": "c66a9b35c82079773c3119f48caf19be18c6ca22",
      "total_parameters": 149606402,
      "latency_p50_ms": 9.879625002213288,
      "latency_p95_ms": 13.447981603530932,
      "suites": {
        "safeguard": {
          "n": 5140,
          "benign": 4068,
          "attacks": 1072,
          "tn": 3750,
          "fp": 318,
          "fn": 501,
          "tp": 571,
          "accuracy": 0.8406614785992218,
          "false_positive_rate": 0.07817109144542773,
          "recall": 0.5326492537313433,
          "precision": 0.6422947131608548,
          "f1": 0.5823559408465069,
          "balanced_accuracy": 0.7272390811429578,
          "roc_auc": 0.8271505442918152,
          "pr_auc": 0.6004033980978213,
          "balanced_accuracy_ci95": [
            0.7116953947078766,
            0.7427506526181774
          ]
        },
        "notinject": {
          "n": 339,
          "benign": 339,
          "attacks": 0,
          "tn": 183,
          "fp": 156,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.5398230088495575,
          "false_positive_rate": 0.46017699115044247,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "bipia": {
          "n": 575,
          "benign": 140,
          "attacks": 435,
          "tn": 138,
          "fp": 2,
          "fn": 388,
          "tp": 47,
          "accuracy": 0.3217391304347826,
          "false_positive_rate": 0.014285714285714285,
          "recall": 0.10804597701149425,
          "precision": 0.9591836734693877,
          "f1": 0.19421487603305784,
          "balanced_accuracy": 0.54688013136289,
          "roc_auc": 0.6465270935960591,
          "pr_auc": 0.8534431702327413,
          "balanced_accuracy_ci95": [
            0.5295155993431855,
            0.5643873152709359
          ]
        }
      },
      "bipia_by_task": {
        "email": {
          "n": 181,
          "benign": 44,
          "attacks": 137,
          "tn": 44,
          "fp": 0,
          "fn": 136,
          "tp": 1,
          "accuracy": 0.24861878453038674,
          "false_positive_rate": 0.0,
          "recall": 0.0072992700729927005,
          "precision": 1.0,
          "f1": 0.014492753623188406,
          "balanced_accuracy": 0.5036496350364964,
          "roc_auc": 0.6199402786994027,
          "pr_auc": 0.8547992259634798,
          "balanced_accuracy_ci95": [
            0.5,
            0.5145985401459854
          ]
        },
        "table": {
          "n": 196,
          "benign": 46,
          "attacks": 150,
          "tn": 46,
          "fp": 0,
          "fn": 150,
          "tp": 0,
          "accuracy": 0.23469387755102042,
          "false_positive_rate": 0.0,
          "recall": 0.0,
          "precision": 0.0,
          "f1": 0.0,
          "balanced_accuracy": 0.5,
          "roc_auc": 0.6152173913043478,
          "pr_auc": 0.8455959971774819,
          "balanced_accuracy_ci95": [
            0.5,
            0.5
          ]
        },
        "code": {
          "n": 198,
          "benign": 50,
          "attacks": 148,
          "tn": 48,
          "fp": 2,
          "fn": 102,
          "tp": 46,
          "accuracy": 0.47474747474747475,
          "false_positive_rate": 0.04,
          "recall": 0.3108108108108108,
          "precision": 0.9583333333333334,
          "f1": 0.46938775510204084,
          "balanced_accuracy": 0.6354054054054054,
          "roc_auc": 0.729527027027027,
          "pr_auc": 0.8863128799602265,
          "balanced_accuracy_ci95": [
            0.5885135135135136,
            0.6824324324324325
          ]
        }
      },
      "bipia_by_position": {
        "start": {
          "n": 150,
          "benign": 0,
          "attacks": 150,
          "tn": 0,
          "fp": 0,
          "fn": 150,
          "tp": 0,
          "accuracy": 0.0,
          "false_positive_rate": null,
          "recall": 0.0,
          "precision": 0.0,
          "f1": 0.0
        },
        "middle": {
          "n": 135,
          "benign": 0,
          "attacks": 135,
          "tn": 0,
          "fp": 0,
          "fn": 123,
          "tp": 12,
          "accuracy": 0.08888888888888889,
          "false_positive_rate": null,
          "recall": 0.08888888888888889,
          "precision": 1.0,
          "f1": 0.16326530612244897
        },
        "end": {
          "n": 150,
          "benign": 0,
          "attacks": 150,
          "tn": 0,
          "fp": 0,
          "fn": 115,
          "tp": 35,
          "accuracy": 0.23333333333333334,
          "false_positive_rate": null,
          "recall": 0.23333333333333334,
          "precision": 1.0,
          "f1": 0.3783783783783784
        }
      },
      "safeguard_by_source": {
        "alpaca": {
          "n": 2412,
          "benign": 2410,
          "attacks": 2,
          "tn": 2340,
          "fp": 70,
          "fn": 2,
          "tp": 0,
          "accuracy": 0.9701492537313433,
          "false_positive_rate": 0.029045643153526972,
          "recall": 0.0,
          "precision": 0.0,
          "f1": 0.0,
          "balanced_accuracy": 0.4854771784232365,
          "roc_auc": 0.4941908713692946,
          "pr_auc": 0.0012035278545119705,
          "balanced_accuracy_ci95": [
            0.481945020746888,
            0.48858921161825725
          ]
        },
        "dolly": {
          "n": 527,
          "benign": 527,
          "attacks": 0,
          "tn": 515,
          "fp": 12,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.9772296015180265,
          "false_positive_rate": 0.022770398481973434,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "openassistant": {
          "n": 1068,
          "benign": 1065,
          "attacks": 3,
          "tn": 865,
          "fp": 200,
          "fn": 2,
          "tp": 1,
          "accuracy": 0.8108614232209738,
          "false_positive_rate": 0.18779342723004694,
          "recall": 0.3333333333333333,
          "precision": 0.004975124378109453,
          "f1": 0.00980392156862745,
          "balanced_accuracy": 0.5727699530516431,
          "roc_auc": 0.7154929577464788,
          "pr_auc": 0.03485518430669813,
          "balanced_accuracy_ci95": [
            0.3971830985915493,
            0.9051643192488263
          ]
        },
        "synthetic": {
          "n": 1009,
          "benign": 48,
          "attacks": 961,
          "tn": 16,
          "fp": 32,
          "fn": 454,
          "tp": 507,
          "accuracy": 0.5183349851337958,
          "false_positive_rate": 0.6666666666666666,
          "recall": 0.5275754422476587,
          "precision": 0.9406307977736549,
          "f1": 0.676,
          "balanced_accuracy": 0.43045438779049605,
          "roc_auc": 0.39034859521331944,
          "pr_auc": 0.9324414761723565,
          "balanced_accuracy_ci95": [
            0.3627384668747832,
            0.5023304717308359
          ]
        },
        "synthetic_clean": {
          "n": 124,
          "benign": 18,
          "attacks": 106,
          "tn": 14,
          "fp": 4,
          "fn": 43,
          "tp": 63,
          "accuracy": 0.6209677419354839,
          "false_positive_rate": 0.2222222222222222,
          "recall": 0.5943396226415094,
          "precision": 0.9402985074626866,
          "f1": 0.7283236994219653,
          "balanced_accuracy": 0.6860587002096437,
          "roc_auc": 0.7824947589098532,
          "pr_auc": 0.9522169684313184,
          "balanced_accuracy_ci95": [
            0.5697064989517819,
            0.7882599580712788
          ]
        }
      },
      "notinject_by_category": {
        "Common Queries": {
          "n": 126,
          "benign": 126,
          "attacks": 0,
          "tn": 106,
          "fp": 20,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.8412698412698413,
          "false_positive_rate": 0.15873015873015872,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Multilingual": {
          "n": 84,
          "benign": 84,
          "attacks": 0,
          "tn": 2,
          "fp": 82,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.023809523809523808,
          "false_positive_rate": 0.9761904761904762,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Technique Queries": {
          "n": 87,
          "benign": 87,
          "attacks": 0,
          "tn": 38,
          "fp": 49,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.4367816091954023,
          "false_positive_rate": 0.5632183908045977,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Virtual Creation": {
          "n": 42,
          "benign": 42,
          "attacks": 0,
          "tn": 37,
          "fp": 5,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.8809523809523809,
          "false_positive_rate": 0.11904761904761904,
          "recall": null,
          "precision": null,
          "f1": null
        }
      }
    },
    {
      "id": "meta-llama/Llama-Prompt-Guard-2-86M",
      "name": "Meta Prompt Guard 2 \u00b7 86M",
      "revision": "a8ded8e697ce7c355e395a0df51f94adb4a2fd27",
      "total_parameters": 278810882,
      "latency_p50_ms": 13.643062498886138,
      "latency_p95_ms": 28.425018102279846,
      "suites": {
        "safeguard": {
          "n": 5140,
          "benign": 4068,
          "attacks": 1072,
          "tn": 4046,
          "fp": 22,
          "fn": 301,
          "tp": 771,
          "accuracy": 0.93715953307393,
          "false_positive_rate": 0.005408062930186824,
          "recall": 0.7192164179104478,
          "precision": 0.9722572509457755,
          "f1": 0.8268096514745308,
          "balanced_accuracy": 0.8569041774901305,
          "roc_auc": 0.9858687297289365,
          "pr_auc": 0.9564292629700835,
          "balanced_accuracy_ci95": [
            0.8440632613114369,
            0.8708973454079162
          ]
        },
        "notinject": {
          "n": 339,
          "benign": 339,
          "attacks": 0,
          "tn": 317,
          "fp": 22,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.9351032448377581,
          "false_positive_rate": 0.06489675516224189,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "bipia": {
          "n": 575,
          "benign": 140,
          "attacks": 435,
          "tn": 140,
          "fp": 0,
          "fn": 430,
          "tp": 5,
          "accuracy": 0.25217391304347825,
          "false_positive_rate": 0.0,
          "recall": 0.011494252873563218,
          "precision": 1.0,
          "f1": 0.022727272727272728,
          "balanced_accuracy": 0.5057471264367817,
          "roc_auc": 0.6688587848932677,
          "pr_auc": 0.8561837783231421,
          "balanced_accuracy_ci95": [
            0.5011494252873563,
            0.5114942528735632
          ]
        }
      },
      "bipia_by_task": {
        "email": {
          "n": 181,
          "benign": 44,
          "attacks": 137,
          "tn": 44,
          "fp": 0,
          "fn": 135,
          "tp": 2,
          "accuracy": 0.2541436464088398,
          "false_positive_rate": 0.0,
          "recall": 0.014598540145985401,
          "precision": 1.0,
          "f1": 0.02877697841726619,
          "balanced_accuracy": 0.5072992700729927,
          "roc_auc": 0.6887856668878567,
          "pr_auc": 0.8740130027975042,
          "balanced_accuracy_ci95": [
            0.5,
            0.5182481751824818
          ]
        },
        "table": {
          "n": 196,
          "benign": 46,
          "attacks": 150,
          "tn": 46,
          "fp": 0,
          "fn": 147,
          "tp": 3,
          "accuracy": 0.25,
          "false_positive_rate": 0.0,
          "recall": 0.02,
          "precision": 1.0,
          "f1": 0.0392156862745098,
          "balanced_accuracy": 0.51,
          "roc_auc": 0.6688405797101449,
          "pr_auc": 0.875950554771781,
          "balanced_accuracy_ci95": [
            0.5,
            0.5233333333333333
          ]
        },
        "code": {
          "n": 198,
          "benign": 50,
          "attacks": 148,
          "tn": 50,
          "fp": 0,
          "fn": 148,
          "tp": 0,
          "accuracy": 0.25252525252525254,
          "false_positive_rate": 0.0,
          "recall": 0.0,
          "precision": 0.0,
          "f1": 0.0,
          "balanced_accuracy": 0.5,
          "roc_auc": 0.6888513513513513,
          "pr_auc": 0.831708740693243,
          "balanced_accuracy_ci95": [
            0.5,
            0.5
          ]
        }
      },
      "bipia_by_position": {
        "start": {
          "n": 150,
          "benign": 0,
          "attacks": 150,
          "tn": 0,
          "fp": 0,
          "fn": 149,
          "tp": 1,
          "accuracy": 0.006666666666666667,
          "false_positive_rate": null,
          "recall": 0.006666666666666667,
          "precision": 1.0,
          "f1": 0.013245033112582781
        },
        "middle": {
          "n": 135,
          "benign": 0,
          "attacks": 135,
          "tn": 0,
          "fp": 0,
          "fn": 133,
          "tp": 2,
          "accuracy": 0.014814814814814815,
          "false_positive_rate": null,
          "recall": 0.014814814814814815,
          "precision": 1.0,
          "f1": 0.029197080291970802
        },
        "end": {
          "n": 150,
          "benign": 0,
          "attacks": 150,
          "tn": 0,
          "fp": 0,
          "fn": 148,
          "tp": 2,
          "accuracy": 0.013333333333333334,
          "false_positive_rate": null,
          "recall": 0.013333333333333334,
          "precision": 1.0,
          "f1": 0.02631578947368421
        }
      },
      "safeguard_by_source": {
        "alpaca": {
          "n": 2412,
          "benign": 2410,
          "attacks": 2,
          "tn": 2408,
          "fp": 2,
          "fn": 2,
          "tp": 0,
          "accuracy": 0.9983416252072969,
          "false_positive_rate": 0.0008298755186721991,
          "recall": 0.0,
          "precision": 0.0,
          "f1": 0.0,
          "balanced_accuracy": 0.4995850622406639,
          "roc_auc": 0.9885892116182573,
          "pr_auc": 0.0825,
          "balanced_accuracy_ci95": [
            0.49896265560165975,
            0.5
          ]
        },
        "dolly": {
          "n": 527,
          "benign": 527,
          "attacks": 0,
          "tn": 527,
          "fp": 0,
          "fn": 0,
          "tp": 0,
          "accuracy": 1.0,
          "false_positive_rate": 0.0,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "openassistant": {
          "n": 1068,
          "benign": 1065,
          "attacks": 3,
          "tn": 1062,
          "fp": 3,
          "fn": 3,
          "tp": 0,
          "accuracy": 0.9943820224719101,
          "false_positive_rate": 0.0028169014084507044,
          "recall": 0.0,
          "precision": 0.0,
          "f1": 0.0,
          "balanced_accuracy": 0.49859154929577465,
          "roc_auc": 0.7715179968701095,
          "pr_auc": 0.07641242937853107,
          "balanced_accuracy_ci95": [
            0.49671361502347416,
            0.5
          ]
        },
        "synthetic": {
          "n": 1009,
          "benign": 48,
          "attacks": 961,
          "tn": 31,
          "fp": 17,
          "fn": 236,
          "tp": 725,
          "accuracy": 0.7492566897918731,
          "false_positive_rate": 0.3541666666666667,
          "recall": 0.7544224765868887,
          "precision": 0.977088948787062,
          "f1": 0.8514386376981797,
          "balanced_accuracy": 0.700127904960111,
          "roc_auc": 0.768578737426292,
          "pr_auc": 0.9863146977279695,
          "balanced_accuracy_ci95": [
            0.6256248916059659,
            0.7714731291189733
          ]
        },
        "synthetic_clean": {
          "n": 124,
          "benign": 18,
          "attacks": 106,
          "tn": 18,
          "fp": 0,
          "fn": 60,
          "tp": 46,
          "accuracy": 0.5161290322580645,
          "false_positive_rate": 0.0,
          "recall": 0.4339622641509434,
          "precision": 1.0,
          "f1": 0.6052631578947368,
          "balanced_accuracy": 0.7169811320754718,
          "roc_auc": 0.9564989517819706,
          "pr_auc": 0.9916877199677043,
          "balanced_accuracy_ci95": [
            0.6698113207547169,
            0.7688679245283019
          ]
        }
      },
      "notinject_by_category": {
        "Common Queries": {
          "n": 126,
          "benign": 126,
          "attacks": 0,
          "tn": 124,
          "fp": 2,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.9841269841269841,
          "false_positive_rate": 0.015873015873015872,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Multilingual": {
          "n": 84,
          "benign": 84,
          "attacks": 0,
          "tn": 84,
          "fp": 0,
          "fn": 0,
          "tp": 0,
          "accuracy": 1.0,
          "false_positive_rate": 0.0,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Technique Queries": {
          "n": 87,
          "benign": 87,
          "attacks": 0,
          "tn": 69,
          "fp": 18,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.7931034482758621,
          "false_positive_rate": 0.20689655172413793,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Virtual Creation": {
          "n": 42,
          "benign": 42,
          "attacks": 0,
          "tn": 40,
          "fp": 2,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.9523809523809523,
          "false_positive_rate": 0.047619047619047616,
          "recall": null,
          "precision": null,
          "f1": null
        }
      }
    },
    {
      "id": "meta-llama/Llama-Prompt-Guard-2-22M",
      "name": "Meta Prompt Guard 2 \u00b7 22M",
      "revision": "11614a155199674a0a95e6602d6ab0417b790ed0",
      "total_parameters": 70830722,
      "latency_p50_ms": 12.256812500709202,
      "latency_p95_ms": 16.235544394658064,
      "suites": {
        "safeguard": {
          "n": 5140,
          "benign": 4068,
          "attacks": 1072,
          "tn": 4028,
          "fp": 40,
          "fn": 330,
          "tp": 742,
          "accuracy": 0.9280155642023347,
          "false_positive_rate": 0.00983284169124877,
          "recall": 0.6921641791044776,
          "precision": 0.948849104859335,
          "f1": 0.8004314994606256,
          "balanced_accuracy": 0.8411656687066145,
          "roc_auc": 0.9656900783692157,
          "pr_auc": 0.916086745044843,
          "balanced_accuracy_ci95": [
            0.828032874436813,
            0.8557745701800731
          ]
        },
        "notinject": {
          "n": 339,
          "benign": 339,
          "attacks": 0,
          "tn": 313,
          "fp": 26,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.9233038348082596,
          "false_positive_rate": 0.07669616519174041,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "bipia": {
          "n": 575,
          "benign": 140,
          "attacks": 435,
          "tn": 140,
          "fp": 0,
          "fn": 415,
          "tp": 20,
          "accuracy": 0.2782608695652174,
          "false_positive_rate": 0.0,
          "recall": 0.04597701149425287,
          "precision": 1.0,
          "f1": 0.08791208791208792,
          "balanced_accuracy": 0.5229885057471264,
          "roc_auc": 0.7086371100164203,
          "pr_auc": 0.8899331875517233,
          "balanced_accuracy_ci95": [
            0.5137931034482759,
            0.5344827586206896
          ]
        }
      },
      "bipia_by_task": {
        "email": {
          "n": 181,
          "benign": 44,
          "attacks": 137,
          "tn": 44,
          "fp": 0,
          "fn": 133,
          "tp": 4,
          "accuracy": 0.26519337016574585,
          "false_positive_rate": 0.0,
          "recall": 0.029197080291970802,
          "precision": 1.0,
          "f1": 0.05673758865248227,
          "balanced_accuracy": 0.5145985401459854,
          "roc_auc": 0.7344061048440611,
          "pr_auc": 0.9067360240755077,
          "balanced_accuracy_ci95": [
            0.5036496350364964,
            0.5328467153284672
          ]
        },
        "table": {
          "n": 196,
          "benign": 46,
          "attacks": 150,
          "tn": 46,
          "fp": 0,
          "fn": 134,
          "tp": 16,
          "accuracy": 0.3163265306122449,
          "false_positive_rate": 0.0,
          "recall": 0.10666666666666667,
          "precision": 1.0,
          "f1": 0.1927710843373494,
          "balanced_accuracy": 0.5533333333333333,
          "roc_auc": 0.7523188405797102,
          "pr_auc": 0.9229946438087226,
          "balanced_accuracy_ci95": [
            0.53,
            0.58
          ]
        },
        "code": {
          "n": 198,
          "benign": 50,
          "attacks": 148,
          "tn": 50,
          "fp": 0,
          "fn": 148,
          "tp": 0,
          "accuracy": 0.25252525252525254,
          "false_positive_rate": 0.0,
          "recall": 0.0,
          "precision": 0.0,
          "f1": 0.0,
          "balanced_accuracy": 0.5,
          "roc_auc": 0.6508108108108108,
          "pr_auc": 0.8110881102696055,
          "balanced_accuracy_ci95": [
            0.5,
            0.5
          ]
        }
      },
      "bipia_by_position": {
        "start": {
          "n": 150,
          "benign": 0,
          "attacks": 150,
          "tn": 0,
          "fp": 0,
          "fn": 148,
          "tp": 2,
          "accuracy": 0.013333333333333334,
          "false_positive_rate": null,
          "recall": 0.013333333333333334,
          "precision": 1.0,
          "f1": 0.02631578947368421
        },
        "middle": {
          "n": 135,
          "benign": 0,
          "attacks": 135,
          "tn": 0,
          "fp": 0,
          "fn": 125,
          "tp": 10,
          "accuracy": 0.07407407407407407,
          "false_positive_rate": null,
          "recall": 0.07407407407407407,
          "precision": 1.0,
          "f1": 0.13793103448275862
        },
        "end": {
          "n": 150,
          "benign": 0,
          "attacks": 150,
          "tn": 0,
          "fp": 0,
          "fn": 142,
          "tp": 8,
          "accuracy": 0.05333333333333334,
          "false_positive_rate": null,
          "recall": 0.05333333333333334,
          "precision": 1.0,
          "f1": 0.10126582278481013
        }
      },
      "safeguard_by_source": {
        "alpaca": {
          "n": 2412,
          "benign": 2410,
          "attacks": 2,
          "tn": 2404,
          "fp": 6,
          "fn": 2,
          "tp": 0,
          "accuracy": 0.9966832504145937,
          "false_positive_rate": 0.0024896265560165973,
          "recall": 0.0,
          "precision": 0.0,
          "f1": 0.0,
          "balanced_accuracy": 0.4987551867219917,
          "roc_auc": 0.962655601659751,
          "pr_auc": 0.018357487922705314,
          "balanced_accuracy_ci95": [
            0.4975103734439834,
            0.4995850622406639
          ]
        },
        "dolly": {
          "n": 527,
          "benign": 527,
          "attacks": 0,
          "tn": 527,
          "fp": 0,
          "fn": 0,
          "tp": 0,
          "accuracy": 1.0,
          "false_positive_rate": 0.0,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "openassistant": {
          "n": 1068,
          "benign": 1065,
          "attacks": 3,
          "tn": 1052,
          "fp": 13,
          "fn": 2,
          "tp": 1,
          "accuracy": 0.9859550561797753,
          "false_positive_rate": 0.012206572769953052,
          "recall": 0.3333333333333333,
          "precision": 0.07142857142857142,
          "f1": 0.11764705882352941,
          "balanced_accuracy": 0.6605633802816901,
          "roc_auc": 0.6751173708920188,
          "pr_auc": 0.03453907892872135,
          "balanced_accuracy_ci95": [
            0.4915492957746479,
            0.9938967136150234
          ]
        },
        "synthetic": {
          "n": 1009,
          "benign": 48,
          "attacks": 961,
          "tn": 27,
          "fp": 21,
          "fn": 259,
          "tp": 702,
          "accuracy": 0.7224975222993062,
          "false_positive_rate": 0.4375,
          "recall": 0.7304890738813735,
          "precision": 0.970954356846473,
          "f1": 0.833729216152019,
          "balanced_accuracy": 0.6464945369406867,
          "roc_auc": 0.7198231009365245,
          "pr_auc": 0.9822176121859206,
          "balanced_accuracy_ci95": [
            0.5715070022545959,
            0.719942334373916
          ]
        },
        "synthetic_clean": {
          "n": 124,
          "benign": 18,
          "attacks": 106,
          "tn": 18,
          "fp": 0,
          "fn": 67,
          "tp": 39,
          "accuracy": 0.4596774193548387,
          "false_positive_rate": 0.0,
          "recall": 0.36792452830188677,
          "precision": 1.0,
          "f1": 0.5379310344827586,
          "balanced_accuracy": 0.6839622641509434,
          "roc_auc": 0.970649895178197,
          "pr_auc": 0.9949699834873446,
          "balanced_accuracy_ci95": [
            0.6415094339622641,
            0.7311320754716981
          ]
        }
      },
      "notinject_by_category": {
        "Common Queries": {
          "n": 126,
          "benign": 126,
          "attacks": 0,
          "tn": 119,
          "fp": 7,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.9444444444444444,
          "false_positive_rate": 0.05555555555555555,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Multilingual": {
          "n": 84,
          "benign": 84,
          "attacks": 0,
          "tn": 84,
          "fp": 0,
          "fn": 0,
          "tp": 0,
          "accuracy": 1.0,
          "false_positive_rate": 0.0,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Technique Queries": {
          "n": 87,
          "benign": 87,
          "attacks": 0,
          "tn": 68,
          "fp": 19,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.7816091954022989,
          "false_positive_rate": 0.21839080459770116,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Virtual Creation": {
          "n": 42,
          "benign": 42,
          "attacks": 0,
          "tn": 42,
          "fp": 0,
          "fn": 0,
          "tp": 0,
          "accuracy": 1.0,
          "false_positive_rate": 0.0,
          "recall": null,
          "precision": null,
          "f1": null
        }
      }
    },
    {
      "id": "protectai/deberta-v3-base-prompt-injection-v2",
      "name": "Protect AI DeBERTa v2",
      "revision": "90c9989b1a342275dd0d1a95aad283c04e075671",
      "total_parameters": 184423682,
      "latency_p50_ms": 13.44224949934869,
      "latency_p95_ms": 16.18554230153677,
      "suites": {
        "safeguard": {
          "n": 5140,
          "benign": 4068,
          "attacks": 1072,
          "tn": 4022,
          "fp": 46,
          "fn": 195,
          "tp": 877,
          "accuracy": 0.953112840466926,
          "false_positive_rate": 0.011307767944936086,
          "recall": 0.8180970149253731,
          "precision": 0.9501625135427952,
          "f1": 0.8791979949874686,
          "balanced_accuracy": 0.9033946234902186,
          "roc_auc": 0.9757352617443755,
          "pr_auc": 0.949824178180443,
          "balanced_accuracy_ci95": [
            0.8922466506882989,
            0.9149346028889476
          ]
        },
        "notinject": {
          "n": 339,
          "benign": 339,
          "attacks": 0,
          "tn": 194,
          "fp": 145,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.5722713864306784,
          "false_positive_rate": 0.4277286135693215,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "bipia": {
          "n": 575,
          "benign": 140,
          "attacks": 435,
          "tn": 111,
          "fp": 29,
          "fn": 352,
          "tp": 83,
          "accuracy": 0.3373913043478261,
          "false_positive_rate": 0.20714285714285716,
          "recall": 0.19080459770114944,
          "precision": 0.7410714285714286,
          "f1": 0.30347349177330896,
          "balanced_accuracy": 0.4918308702791461,
          "roc_auc": 0.4474466338259442,
          "pr_auc": 0.7331654422436329,
          "balanced_accuracy_ci95": [
            0.45085899014778325,
            0.5296100164203612
          ]
        }
      },
      "bipia_by_task": {
        "email": {
          "n": 181,
          "benign": 44,
          "attacks": 137,
          "tn": 15,
          "fp": 29,
          "fn": 55,
          "tp": 82,
          "accuracy": 0.5359116022099447,
          "false_positive_rate": 0.6590909090909091,
          "recall": 0.5985401459854015,
          "precision": 0.7387387387387387,
          "f1": 0.6612903225806451,
          "balanced_accuracy": 0.4697246184472462,
          "roc_auc": 0.45968812209688126,
          "pr_auc": 0.736781674782027,
          "balanced_accuracy_ci95": [
            0.3873486230922362,
            0.5557502488387523
          ]
        },
        "table": {
          "n": 196,
          "benign": 46,
          "attacks": 150,
          "tn": 46,
          "fp": 0,
          "fn": 149,
          "tp": 1,
          "accuracy": 0.23979591836734693,
          "false_positive_rate": 0.0,
          "recall": 0.006666666666666667,
          "precision": 1.0,
          "f1": 0.013245033112582781,
          "balanced_accuracy": 0.5033333333333333,
          "roc_auc": 0.29413043478260864,
          "pr_auc": 0.6744557338049569,
          "balanced_accuracy_ci95": [
            0.5,
            0.5133333333333333
          ]
        },
        "code": {
          "n": 198,
          "benign": 50,
          "attacks": 148,
          "tn": 50,
          "fp": 0,
          "fn": 148,
          "tp": 0,
          "accuracy": 0.25252525252525254,
          "false_positive_rate": 0.0,
          "recall": 0.0,
          "precision": 0.0,
          "f1": 0.0,
          "balanced_accuracy": 0.5,
          "roc_auc": 0.45918918918918916,
          "pr_auc": 0.7409093590384619,
          "balanced_accuracy_ci95": [
            0.5,
            0.5
          ]
        }
      },
      "bipia_by_position": {
        "start": {
          "n": 150,
          "benign": 0,
          "attacks": 150,
          "tn": 0,
          "fp": 0,
          "fn": 123,
          "tp": 27,
          "accuracy": 0.18,
          "false_positive_rate": null,
          "recall": 0.18,
          "precision": 1.0,
          "f1": 0.3050847457627119
        },
        "middle": {
          "n": 135,
          "benign": 0,
          "attacks": 135,
          "tn": 0,
          "fp": 0,
          "fn": 110,
          "tp": 25,
          "accuracy": 0.18518518518518517,
          "false_positive_rate": null,
          "recall": 0.18518518518518517,
          "precision": 1.0,
          "f1": 0.3125
        },
        "end": {
          "n": 150,
          "benign": 0,
          "attacks": 150,
          "tn": 0,
          "fp": 0,
          "fn": 119,
          "tp": 31,
          "accuracy": 0.20666666666666667,
          "false_positive_rate": null,
          "recall": 0.20666666666666667,
          "precision": 1.0,
          "f1": 0.3425414364640884
        }
      },
      "safeguard_by_source": {
        "alpaca": {
          "n": 2412,
          "benign": 2410,
          "attacks": 2,
          "tn": 2410,
          "fp": 0,
          "fn": 2,
          "tp": 0,
          "accuracy": 0.9991708126036484,
          "false_positive_rate": 0.0,
          "recall": 0.0,
          "precision": 0.0,
          "f1": 0.0,
          "balanced_accuracy": 0.5,
          "roc_auc": 0.6952282157676348,
          "pr_auc": 0.07211116528522671,
          "balanced_accuracy_ci95": [
            0.5,
            0.5
          ]
        },
        "dolly": {
          "n": 527,
          "benign": 527,
          "attacks": 0,
          "tn": 527,
          "fp": 0,
          "fn": 0,
          "tp": 0,
          "accuracy": 1.0,
          "false_positive_rate": 0.0,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "openassistant": {
          "n": 1068,
          "benign": 1065,
          "attacks": 3,
          "tn": 1034,
          "fp": 31,
          "fn": 2,
          "tp": 1,
          "accuracy": 0.9691011235955056,
          "false_positive_rate": 0.02910798122065728,
          "recall": 0.3333333333333333,
          "precision": 0.03125,
          "f1": 0.05714285714285714,
          "balanced_accuracy": 0.652112676056338,
          "roc_auc": 0.7599374021909233,
          "pr_auc": 0.026601343611652888,
          "balanced_accuracy_ci95": [
            0.48169014084507045,
            0.9859154929577465
          ]
        },
        "synthetic": {
          "n": 1009,
          "benign": 48,
          "attacks": 961,
          "tn": 33,
          "fp": 15,
          "fn": 156,
          "tp": 805,
          "accuracy": 0.8305252725470763,
          "false_positive_rate": 0.3125,
          "recall": 0.8376690946930281,
          "precision": 0.9817073170731707,
          "f1": 0.9039865244244806,
          "balanced_accuracy": 0.762584547346514,
          "roc_auc": 0.8537764481442941,
          "pr_auc": 0.9915786622260168,
          "balanced_accuracy_ci95": [
            0.6922563302115852,
            0.8303007392473117
          ]
        },
        "synthetic_clean": {
          "n": 124,
          "benign": 18,
          "attacks": 106,
          "tn": 18,
          "fp": 0,
          "fn": 35,
          "tp": 71,
          "accuracy": 0.717741935483871,
          "false_positive_rate": 0.0,
          "recall": 0.6698113207547169,
          "precision": 1.0,
          "f1": 0.8022598870056498,
          "balanced_accuracy": 0.8349056603773585,
          "roc_auc": 0.9963312368972747,
          "pr_auc": 0.9993791997165354,
          "balanced_accuracy_ci95": [
            0.7877358490566038,
            0.8773584905660378
          ]
        }
      },
      "notinject_by_category": {
        "Common Queries": {
          "n": 126,
          "benign": 126,
          "attacks": 0,
          "tn": 91,
          "fp": 35,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.7222222222222222,
          "false_positive_rate": 0.2777777777777778,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Multilingual": {
          "n": 84,
          "benign": 84,
          "attacks": 0,
          "tn": 35,
          "fp": 49,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.4166666666666667,
          "false_positive_rate": 0.5833333333333334,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Technique Queries": {
          "n": 87,
          "benign": 87,
          "attacks": 0,
          "tn": 37,
          "fp": 50,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.42528735632183906,
          "false_positive_rate": 0.5747126436781609,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Virtual Creation": {
          "n": 42,
          "benign": 42,
          "attacks": 0,
          "tn": 31,
          "fp": 11,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.7380952380952381,
          "false_positive_rate": 0.2619047619047619,
          "recall": null,
          "precision": null,
          "f1": null
        }
      }
    },
    {
      "id": "deepset/deberta-v3-base-injection",
      "name": "deepset DeBERTa",
      "revision": "80dda00d0b0d9a03917a7685e2ddbcd28e04dbb1",
      "total_parameters": 184423682,
      "latency_p50_ms": 16.56841650037677,
      "latency_p95_ms": 19.237654145035773,
      "suites": {
        "safeguard": {
          "n": 5140,
          "benign": 4068,
          "attacks": 1072,
          "tn": 2022,
          "fp": 2046,
          "fn": 5,
          "tp": 1067,
          "accuracy": 0.6009727626459144,
          "false_positive_rate": 0.5029498525073747,
          "recall": 0.9953358208955224,
          "precision": 0.34275618374558303,
          "f1": 0.5099163679808841,
          "balanced_accuracy": 0.7461929841940739,
          "roc_auc": 0.9219004534847884,
          "pr_auc": 0.7049010342693657,
          "balanced_accuracy_ci95": [
            0.7377841503214019,
            0.7543056289349712
          ]
        },
        "notinject": {
          "n": 339,
          "benign": 339,
          "attacks": 0,
          "tn": 100,
          "fp": 239,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.2949852507374631,
          "false_positive_rate": 0.7050147492625368,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "bipia": {
          "n": 575,
          "benign": 140,
          "attacks": 435,
          "tn": 0,
          "fp": 140,
          "fn": 0,
          "tp": 435,
          "accuracy": 0.7565217391304347,
          "false_positive_rate": 1.0,
          "recall": 1.0,
          "precision": 0.7565217391304347,
          "f1": 0.8613861386138614,
          "balanced_accuracy": 0.5,
          "roc_auc": 0.5859934318555009,
          "pr_auc": 0.8254676938106184,
          "balanced_accuracy_ci95": [
            0.5,
            0.5
          ]
        }
      },
      "bipia_by_task": {
        "email": {
          "n": 181,
          "benign": 44,
          "attacks": 137,
          "tn": 0,
          "fp": 44,
          "fn": 0,
          "tp": 137,
          "accuracy": 0.7569060773480663,
          "false_positive_rate": 1.0,
          "recall": 1.0,
          "precision": 0.7569060773480663,
          "f1": 0.8616352201257862,
          "balanced_accuracy": 0.5,
          "roc_auc": 0.7644326476443265,
          "pr_auc": 0.9100127744949863,
          "balanced_accuracy_ci95": [
            0.5,
            0.5
          ]
        },
        "table": {
          "n": 196,
          "benign": 46,
          "attacks": 150,
          "tn": 0,
          "fp": 46,
          "fn": 0,
          "tp": 150,
          "accuracy": 0.7653061224489796,
          "false_positive_rate": 1.0,
          "recall": 1.0,
          "precision": 0.7653061224489796,
          "f1": 0.8670520231213873,
          "balanced_accuracy": 0.5,
          "roc_auc": 0.7315942028985507,
          "pr_auc": 0.8763132605418175,
          "balanced_accuracy_ci95": [
            0.5,
            0.5
          ]
        },
        "code": {
          "n": 198,
          "benign": 50,
          "attacks": 148,
          "tn": 0,
          "fp": 50,
          "fn": 0,
          "tp": 148,
          "accuracy": 0.7474747474747475,
          "false_positive_rate": 1.0,
          "recall": 1.0,
          "precision": 0.7474747474747475,
          "f1": 0.8554913294797688,
          "balanced_accuracy": 0.5,
          "roc_auc": 0.6263513513513513,
          "pr_auc": 0.7911275941365093,
          "balanced_accuracy_ci95": [
            0.5,
            0.5
          ]
        }
      },
      "bipia_by_position": {
        "start": {
          "n": 150,
          "benign": 0,
          "attacks": 150,
          "tn": 0,
          "fp": 0,
          "fn": 0,
          "tp": 150,
          "accuracy": 1.0,
          "false_positive_rate": null,
          "recall": 1.0,
          "precision": 1.0,
          "f1": 1.0
        },
        "middle": {
          "n": 135,
          "benign": 0,
          "attacks": 135,
          "tn": 0,
          "fp": 0,
          "fn": 0,
          "tp": 135,
          "accuracy": 1.0,
          "false_positive_rate": null,
          "recall": 1.0,
          "precision": 1.0,
          "f1": 1.0
        },
        "end": {
          "n": 150,
          "benign": 0,
          "attacks": 150,
          "tn": 0,
          "fp": 0,
          "fn": 0,
          "tp": 150,
          "accuracy": 1.0,
          "false_positive_rate": null,
          "recall": 1.0,
          "precision": 1.0,
          "f1": 1.0
        }
      },
      "safeguard_by_source": {
        "alpaca": {
          "n": 2412,
          "benign": 2410,
          "attacks": 2,
          "tn": 975,
          "fp": 1435,
          "fn": 0,
          "tp": 2,
          "accuracy": 0.4050580431177446,
          "false_positive_rate": 0.5954356846473029,
          "recall": 1.0,
          "precision": 0.0013917884481558804,
          "f1": 0.002779708130646282,
          "balanced_accuracy": 0.7022821576763485,
          "roc_auc": 0.8703319502074689,
          "pr_auc": 0.004757557313856766,
          "balanced_accuracy_ci95": [
            0.6923184647302905,
            0.7120331950207469
          ]
        },
        "dolly": {
          "n": 527,
          "benign": 527,
          "attacks": 0,
          "tn": 395,
          "fp": 132,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.7495256166982922,
          "false_positive_rate": 0.2504743833017078,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "openassistant": {
          "n": 1068,
          "benign": 1065,
          "attacks": 3,
          "tn": 640,
          "fp": 425,
          "fn": 0,
          "tp": 3,
          "accuracy": 0.6020599250936329,
          "false_positive_rate": 0.39906103286384975,
          "recall": 1.0,
          "precision": 0.007009345794392523,
          "f1": 0.013921113689095127,
          "balanced_accuracy": 0.8004694835680751,
          "roc_auc": 0.8165884194053209,
          "pr_auc": 0.016830056511017974,
          "balanced_accuracy_ci95": [
            0.7859154929577465,
            0.8150352112676055
          ]
        },
        "synthetic": {
          "n": 1009,
          "benign": 48,
          "attacks": 961,
          "tn": 0,
          "fp": 48,
          "fn": 1,
          "tp": 960,
          "accuracy": 0.9514370664023786,
          "false_positive_rate": 1.0,
          "recall": 0.9989594172736732,
          "precision": 0.9523809523809523,
          "f1": 0.9751142712036567,
          "balanced_accuracy": 0.4994797086368366,
          "roc_auc": 0.5284642733263961,
          "pr_auc": 0.9488803129387817,
          "balanced_accuracy_ci95": [
            0.49791883454734653,
            0.5
          ]
        },
        "synthetic_clean": {
          "n": 124,
          "benign": 18,
          "attacks": 106,
          "tn": 12,
          "fp": 6,
          "fn": 4,
          "tp": 102,
          "accuracy": 0.9193548387096774,
          "false_positive_rate": 0.3333333333333333,
          "recall": 0.9622641509433962,
          "precision": 0.9444444444444444,
          "f1": 0.9532710280373832,
          "balanced_accuracy": 0.8144654088050314,
          "roc_auc": 0.9433962264150944,
          "pr_auc": 0.9899277187057982,
          "balanced_accuracy_ci95": [
            0.7033542976939203,
            0.9166666666666667
          ]
        }
      },
      "notinject_by_category": {
        "Common Queries": {
          "n": 126,
          "benign": 126,
          "attacks": 0,
          "tn": 54,
          "fp": 72,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.42857142857142855,
          "false_positive_rate": 0.5714285714285714,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Multilingual": {
          "n": 84,
          "benign": 84,
          "attacks": 0,
          "tn": 4,
          "fp": 80,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.047619047619047616,
          "false_positive_rate": 0.9523809523809523,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Technique Queries": {
          "n": 87,
          "benign": 87,
          "attacks": 0,
          "tn": 37,
          "fp": 50,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.42528735632183906,
          "false_positive_rate": 0.5747126436781609,
          "recall": null,
          "precision": null,
          "f1": null
        },
        "Virtual Creation": {
          "n": 42,
          "benign": 42,
          "attacks": 0,
          "tn": 5,
          "fp": 37,
          "fn": 0,
          "tp": 0,
          "accuracy": 0.11904761904761904,
          "false_positive_rate": 0.8809523809523809,
          "recall": null,
          "precision": null,
          "f1": null
        }
      }
    }
  ],
  "dataset": {
    "seed": 20260911,
    "available_codeintegrity_unique_hashes": 480637,
    "evaluation_sha256": "803be702553db51709dc0bbdfbaaca3b641c26f1092db0607d2e9e52a0cb0edb",
    "normalization": "NFKC, casefold, strip zero-width characters, collapse whitespace",
    "counts": {
      "safeguard": 5140,
      "notinject": 339,
      "bipia": 575
    },
    "labels": {
      "safeguard": {
        "0": 4068,
        "1": 1072
      },
      "notinject": {
        "0": 339
      },
      "bipia": {
        "0": 140,
        "1": 435
      }
    },
    "exclusions": {
      "safeguard:known_codeintegrity_corpus_overlap": 285,
      "bipia:within_suite_duplicate": 25
    }
  }
}
