Skip to content
KitploitKITPLOIT
ToolsExploitsBlog
Log in
Submit
ToolsExploitsBlog
Submit

Hacking, PenTest, and Cybersecurity Tools for Your Security Arsenal!

Kitploit is a directory of hacking, cybersecurity, and pentesting tools. Discover the latest project updates to find vulnerabilities, analyze systems, automate testing, and strengthen your security.

FeedsContactPrivacy© 2026 Kitploit

Tool Directory

Categories

View all categories
Loading categories
CyBiasBench — Benchmarking Bias in LLM Agents for Cyber-Attack Scenarios | Kitploit
Tools/GitHubGitHub/harry24k/cybiasbench
Vulnerability AnalysisWeb Application ExploitationPenetration TestingMachine LearningPapers & ResearchLearning & EducationCurated ResourcesAI SecurityLabs & Practice
GitHubharry24k/cybiasbench

CyBiasBench

525 days agoNot yet reviewed

Benchmarking Bias in LLM Agents for Cyber-Attack Scenarios

View Repository
Share

CyBiasBench

A benchmark for behavioral bias in LLM offensive-security agents. Each record summarizes one agent's session against one intentionally vulnerable target application, capturing not just whether attacks succeeded but how the agent chose, sequenced, and stuck with attack families.

Status: public release accompanying the NeurIPS 2026 Evaluations & Datasets Track paper. Paper: https://arxiv.org/abs/2605.07830. Interactive result dashboard: https://trustworthyai.co.kr/CyBiasBench/. Dataset on Hugging Face: https://huggingface.co/datasets/taeng0204/CyBiasBench.

What's in the box

  • 5 agents — claude, codex, gemini, kimi, glm
  • 3 vulnerable target apps — juice-shop, mlflow, vuln-shop
  • Two settings:
    • Bias observation (free-choice): 4 prompt conditions × 3 reps = 180 sessions. Conditions cross prompt guidance (guided / unguided) with output structure (structured / unstructured).
    • Bias injection (requested-family): 10 attack families × 3 reps = 450 sessions. Each session is asked to focus on one specific family (SQLi, CMDi, SSRF, IDOR, …).
  • 11 columns per row in the bias observation setting (counts, session ASR, entropy, most-selected family, Selection CR1, total tokens). 13 columns per row in the bias injection setting (compliance, requested-family ASR, same counts).
  • Long-form companion tables with per-(session, attack family) breakdowns.
  • Sanitized request-level traces (paper Table 5 schema): 19,116 rows (observation) + 258,019 rows (injection) of HTTP requests with attack family labels, classifier rule identifiers, CAPEC / CWE mappings, and success flags plus per-session request order. Bodies, headers, absolute timestamps, and fine-grained timing are stripped.

The paper analyzes 10 attack families (sqli, xss, cmdi, path_traversal, auth_bypass, idor, ssrf, csrf, file_upload, info_disclosure) chosen via CAPEC + CWE + OWASP WSTG analysis as the families most directly testable against the three target applications. The trace-level classifier inherits an eleventh label (deserialization) from OWASP CRS rule family 944; it is never a requested target and accounts for <0.1% of trace rows. See schema/glossary.md §"Attack-family scope" for full details.

Total release: 630 aggregated records + 277,135 trace rows ≈ 48 MB.

Interpretation and limits

These measurements describe five deployed-agent snapshots on three web targets, not stable model rankings. Allocation reflects the model, system prompt, CLI, tools, and target interaction. Vulnerability availability, discoverability, execution difficulty, and request costs also affect family shares; the design does not identify intrinsic preference or separate these factors causally.

Session-bootstrap analysis under the original automatic labels finds no improvement in aggregate mean session ASR under steering. It does not establish a reliable decrease for every agent. Alternative request-counting units preserve between-agent separation, while changing its magnitude. Decision value at larger scale, superiority to simple diversity heuristics, and generalization to unseen targets remain open. See HOW_TO_REPRODUCE.md for the reproducible analyses.

Quick start

git clone https://github.com/Harry24k/CyBiasBench.git
cd CyBiasBench
pip install -e .          # loader: pandas + pyarrow
pip install -e '.[jsd]'   # add matplotlib + scipy + numpy for re-analysis

python -c "
from cybiasbench import load_bias_observation, load_bias_injection
obs = load_bias_observation()
inj = load_bias_injection()
print(obs.wide.shape, inj.wide.shape)               # (180, 11) (450, 13)
print(obs.wide.groupby('agent').session_asr.mean()) # mean session ASR per agent
"

Or load CSV / Parquet directly:

import pandas as pd
obs = pd.read_parquet("data/bias_observation_records.parquet")

The same release is also available on Hugging Face. With datasets installed (pip install datasets):

from datasets import load_dataset
obs = load_dataset("taeng0204/CyBiasBench", "bias_observation_records",
                   split="test", revision="v1.0.0")

Each of the six tables is a separate configuration. The test split contains the complete table; it does not denote a held-out partition.

Reproduction: see HOW_TO_REPRODUCE.md for the two-tier reproduction guide (R1 re-analysis from the released traces; R3 full re-experiment via benchmark/).

Layout

CyBiasBench/
├── data/                                 # tabular release (CSV + Parquet)
│   ├── bias_observation_records.{csv,parquet}        180 rows
│   ├── bias_observation_family_long.{csv,parquet}    1,082 rows
│   ├── bias_observation_traces.{csv,parquet}         19,116 rows (request-level)
│   ├── bias_injection_records.{csv,parquet}          450 rows
│   ├── bias_injection_family_long.{csv,parquet}      1,597 rows
│   └── bias_injection_traces.{csv,parquet}           258,019 rows (request-level)
├── schema/
│   ├── glossary.md                       metric definitions
│   ├── columns_bias_observation.md
│   ├── columns_bias_injection.md
│   └── columns_traces.md
├── src/cybiasbench/                      python loader package
├── scripts/                              R1 re-analysis + R3 post-processing
│   ├── build_records.py                  final_stats JSON → aggregate CSV/Parquet
│   ├── build_traces.py                   labeled JSONL → request-level CSV/Parquet
│   ├── verify_consistency.py             sanity suite (+ --round-trip closure)
│   ├── reviewer_revisions.py             bootstrap CIs + request-unit robustness
│   ├── momentum_repro.py                 compliance correlations + estimator comparison
│   ├── agent_identification.py           random-forest agent identification
│   ├── temporal_repro.py                 response to failure in request order
│   ├── classify/                         OWASP CRS + victim-specific classifiers
│   ├── verify/                           response heuristics + session verifier
│   ├── aggregate/                        final_stats.json generator
│   └── jsd_repro/                        Prompt-stability JSD reproduction
│       ├── table2_jsd_robustness.py      paper Table 2 (within/between-agent JSD)
│       └── prompt_profile_stability.py   paper appendix per-condition JSD
├── benchmark/                            R3 re-experiment harness (Docker + agents)
│   ├── run.sh                            one invocation = one (agent, target, prompt) session
│   ├── agents/                           6 agent CLI Dockerfiles (CLI versions pinned)
│   ├── metrics/                          LiteLLM proxy + mitmproxy HTTP logger
│   ├── prompts/                          guided/unguided + 10 injection prompts
│   ├── output_formats/                   4 condition templates
│   └── victims/                          locally-built target images
├── HOW_TO_REPRODUCE.md                   R1 + R3 reproduction walkthrough
├── DATASHEET.md                          Gebru et al. datasheet
├── RESPONSIBLE_USE.md                    dual-use guidance
├── croissant.json                        Croissant metadata (core + RAI)
├── LICENSE                               CC BY-NC-ND 4.0 (data / docs)
├── LICENSE-CODE                          MIT (code)
└── dashboard/                            Next.js source of the interactive result dashboard

Settings (matches paper terminology)

The two settings here match Sections 4 and 5 of the accompanying paper:

SettingPaper sectionRecordsDesign
Bias observation§4 ("free-choice setting")1805 agents × 3 targets × 4 conditions × 3 reps
Bias injection§54505 agents × 3 targets × 10 requested families × 3 reps
Download Tool

Most Popular

View all →

Discover the most used tools by our community.

  • 1

    OsintHunter

    letox4/osinthunter

    🎯 The ultimate OSINT toolkit: investigation graph, confidence scoring, alias families, map, timeline and reports. 49 modules, 400+ sites, no API keys required.

    546
  • 2

    ghostlock-app

    yukonga/ghostlock-app

    GhostLock One-Tap Execution App (CVE-2026-43499)

    477
  • 3

    ARTEX

    baarai/artex

    LLM multi-agent autonomous penetration testing platform with planner/worker agents, asset and exploration graphs, MITM traffic recording, and vulnerability retesting.

    377
  • 4

    redforge

    cybersecurity91/redforge

    Browser-side red team toolkit with 102 offline tools for recon, payload generation, credential attacks, crypto, encoding and engagement reporting.

    324
  • 5

    estorides

    grisuno/estorides

    Aggregates 99+ free OSINT sources in parallel, resolves entities into a knowledge graph, and exports STIX 2.1, MISP, or GraphML with LLM-assisted analysis.

    269
  • 6

    CyberStrikeAI

    ed1s0nz/cyberstrikeai

    The system of action for AI-native cybersecurity—where intent becomes governed execution, evidence becomes operational memory, and every operation improves the next.

    252

Explore all tools

Browse our collection of tools

View all tools →