{
  "ruleset": "2026.10.1",
  "generated_at": "2026-10-21T21:40:39.613Z",
  "thresholds": {
    "suspicious": 0.35,
    "malicious": 0.7
  },
  "rows": [
    {
      "dataset": "deepset/prompt-injections",
      "split": "train",
      "role": "dev",
      "threshold": "flagged",
      "n": 546,
      "tp": 120,
      "fp": 0,
      "tn": 343,
      "fn": 83,
      "precision": 1,
      "recall": 0.591,
      "f1": 0.743,
      "fpr": 0,
      "avg_ms": 0.187
    },
    {
      "dataset": "deepset/prompt-injections",
      "split": "train",
      "role": "dev",
      "threshold": "malicious",
      "n": 546,
      "tp": 49,
      "fp": 0,
      "tn": 343,
      "fn": 154,
      "precision": 1,
      "recall": 0.241,
      "f1": 0.389,
      "fpr": 0,
      "avg_ms": 0.187
    },
    {
      "dataset": "deepset/prompt-injections",
      "split": "test",
      "role": "held-out",
      "threshold": "flagged",
      "n": 116,
      "tp": 16,
      "fp": 0,
      "tn": 56,
      "fn": 44,
      "precision": 1,
      "recall": 0.267,
      "f1": 0.421,
      "fpr": 0,
      "avg_ms": 0.089
    },
    {
      "dataset": "deepset/prompt-injections",
      "split": "test",
      "role": "held-out",
      "threshold": "malicious",
      "n": 116,
      "tp": 10,
      "fp": 0,
      "tn": 56,
      "fn": 50,
      "precision": 1,
      "recall": 0.167,
      "f1": 0.286,
      "fpr": 0,
      "avg_ms": 0.089
    },
    {
      "dataset": "jackhhao/jailbreak-classification",
      "split": "train",
      "role": "dev",
      "threshold": "flagged",
      "n": 1044,
      "tp": 407,
      "fp": 0,
      "tn": 517,
      "fn": 120,
      "precision": 1,
      "recall": 0.772,
      "f1": 0.872,
      "fpr": 0,
      "avg_ms": 0.587
    },
    {
      "dataset": "jackhhao/jailbreak-classification",
      "split": "train",
      "role": "dev",
      "threshold": "malicious",
      "n": 1044,
      "tp": 271,
      "fp": 0,
      "tn": 517,
      "fn": 256,
      "precision": 1,
      "recall": 0.514,
      "f1": 0.679,
      "fpr": 0,
      "avg_ms": 0.587
    },
    {
      "dataset": "jackhhao/jailbreak-classification",
      "split": "test",
      "role": "held-out",
      "threshold": "flagged",
      "n": 262,
      "tp": 112,
      "fp": 1,
      "tn": 122,
      "fn": 27,
      "precision": 0.991,
      "recall": 0.806,
      "f1": 0.889,
      "fpr": 0.008,
      "avg_ms": 0.66
    },
    {
      "dataset": "jackhhao/jailbreak-classification",
      "split": "test",
      "role": "held-out",
      "threshold": "malicious",
      "n": 262,
      "tp": 81,
      "fp": 0,
      "tn": 123,
      "fn": 58,
      "precision": 1,
      "recall": 0.583,
      "f1": 0.736,
      "fpr": 0,
      "avg_ms": 0.66
    },
    {
      "dataset": "Lakera/gandalf_ignore_instructions",
      "split": "test",
      "role": "held-out",
      "threshold": "flagged",
      "n": 112,
      "tp": 74,
      "fp": 0,
      "tn": 0,
      "fn": 38,
      "precision": 1,
      "recall": 0.661,
      "f1": 0.796,
      "fpr": null,
      "avg_ms": 0.073
    },
    {
      "dataset": "Lakera/gandalf_ignore_instructions",
      "split": "test",
      "role": "held-out",
      "threshold": "malicious",
      "n": 112,
      "tp": 68,
      "fp": 0,
      "tn": 0,
      "fn": 44,
      "precision": 1,
      "recall": 0.607,
      "f1": 0.756,
      "fpr": null,
      "avg_ms": 0.073
    },
    {
      "dataset": "benign: Wikipedia (300 random)",
      "split": "-",
      "role": "held-out",
      "threshold": "flagged",
      "n": 300,
      "tp": 0,
      "fp": 4,
      "tn": 296,
      "fn": 0,
      "precision": 0,
      "recall": null,
      "f1": null,
      "fpr": 0.013,
      "avg_ms": 10.477
    },
    {
      "dataset": "benign: Wikipedia (300 random)",
      "split": "-",
      "role": "held-out",
      "threshold": "malicious",
      "n": 300,
      "tp": 0,
      "fp": 0,
      "tn": 300,
      "fn": 0,
      "precision": null,
      "recall": null,
      "f1": null,
      "fpr": 0,
      "avg_ms": 10.477
    },
    {
      "dataset": "benign: GitHub READMEs (150)",
      "split": "-",
      "role": "held-out",
      "threshold": "flagged",
      "n": 144,
      "tp": 0,
      "fp": 8,
      "tn": 136,
      "fn": 0,
      "precision": 0,
      "recall": null,
      "f1": null,
      "fpr": 0.056,
      "avg_ms": 16.103
    },
    {
      "dataset": "benign: GitHub READMEs (150)",
      "split": "-",
      "role": "held-out",
      "threshold": "malicious",
      "n": 144,
      "tp": 0,
      "fp": 4,
      "tn": 140,
      "fn": 0,
      "precision": 0,
      "recall": null,
      "f1": null,
      "fpr": 0.028,
      "avg_ms": 16.103
    }
  ]
}

agentgate-source: code mirror on Cloudflare Pages. All repos