README.md

metadata

license: cc-by-nc-4.0
task_categories:
  - tabular-classification
  - time-series-forecasting
tags:
  - cybersecurity
  - adversarial-machine-learning
  - ai-security
  - adversarial-attacks
  - evasion-attacks
  - apt
  - synthetic-data
  - ml-security
  - model-robustness
  - mlops-security
pretty_name: CYB011 — Synthetic AI Evasion Attack Trajectories (Sample)
size_categories:
  - 10K<n<100K

CYB011 — Synthetic AI Evasion Attack Trajectory Dataset (Sample)

XpertSystems.ai Synthetic Data Platform · SKU: CYB011-SAMPLE · Version 1.0.0

This is a free preview of the full CYB011 — Synthetic AI Evasion Attack Trajectory Dataset product. It contains roughly ~4% of the full dataset at identical schema, attacker-tier distribution, and statistical fingerprint, so you can evaluate fit before licensing the full product.

🤖 Trained baseline + leakage diagnostic available: xpertsystems/cyb011-baseline-classifier — XGBoost + PyTorch MLP for 7-class adversarial attack phase classification (the dataset's headline target), group-aware split by campaign_id, multi-seed evaluation (acc 0.867 ± 0.010, ROC-AUC 0.977 ± 0.002). Includes a comprehensive leakage_diagnostic.json documenting 6 oracle paths discovered across the dataset's targets, 4 README-suggested headline targets that are unlearnable on the sample after honest leak removal, and the missing nation_state attacker tier. Buyers planning adversarial ML research should read the diagnostic first.

File	Rows (sample)	Rows (full)	Description
network_topology.csv	~200	~2,800	Network segment / defender registry
campaign_summary.csv	~200	~5,500	Per-campaign aggregate outcomes
campaign_events.csv	~13,310	~55,000	Discrete campaign event log
attack_trajectories.csv	~14,000	~320,000	Per-timestep adversarial trajectories

Dataset Summary

CYB011 simulates end-to-end adversarial AI evasion attack campaigns against ML-based security detection systems, modeled as a 6-phase adversarial state machine (data has 7 phases — adds idle_dwell):

• 6 adversarial phases: reconnaissance → feature_space_probe → perturbation_craft → evasion_attempt → feedback_adaptation → campaign_consolidation
• 4 attacker capability tiers: script_kiddie, opportunistic, advanced_persistent_threat (APT), nation_state — with per-tier ε-budgets (L∞ perturbation), query budgets (50 → 5,000), base evasion rates, and stealth weights
• 8 defender detection architectures with per-architecture detection_strength (e.g. ensemble_layered 0.91, gradient_boosted 0.78, neural_network 0.74, isolation_forest 0.62)
• L∞ perturbation budget modeling — calibrated mean ε ≈ 0.185 representing realistic imperceptibility constraints
• Query budget tracking — black-box vs white-box attack distinction
• Concept drift injection — adversarial data poisoning of training distributions, ~8% injection rate
• Retraining trigger modeling — defender model refresh after drift detection (~14% trigger rate)
• Transfer attack modeling — perturbations crafted on surrogate models, 31% transfer success rate
• Honeypot density — deception model coverage (5% baseline)
• Coordinated multi-attacker campaigns with 12% coordination rate
• MLOps security signals — gradient access patterns, feature-space probing, lateral pivoting between models

Trained Baseline + Leakage Audit Available

A working baseline classifier + comprehensive leakage diagnostic is published at xpertsystems/cyb011-baseline-classifier.

Component	Detail
Primary task	7-class attack_phase classification (the dataset's headline target)
Secondary artifact	leakage_diagnostic.json — 6 oracle paths + 4 unlearnable targets + missing tier note
Models	XGBoost (model_xgb.json) + PyTorch MLP (model_mlp.safetensors)
Features	37 (after one-hot encoding); pipeline included as feature_engineering.py
Split	Group-aware (GroupShuffleSplit on campaign_id) — 200 campaigns, ~30 in test fold
Validation	Single seed + multi-seed aggregate across 10 seeds
Demo	inference_example.ipynb — end-to-end copy-paste
Headline metrics	XGBoost: accuracy 0.867 ± 0.010, macro ROC-AUC 0.977 ± 0.002 (multi-seed)

Important findings for buyers planning CYB011 ML work (full detail in leakage_diagnostic.json):

Missing nation_state attacker tier: README lists 4 tiers; sample contains only 3 (script_kiddie 50%, opportunistic 40%, APT 10%). Nation_state events are entirely absent. Models trained on this sample cannot generalize to nation_state actors.

6 oracle paths documented across the dataset's targets:

Phase target oracles (3 paths — must be dropped):

1. detection_outcome (!= suppressed_alert → 100% evasion_attempt)
2. detector_confidence_score (threshold-derived from detection_outcome)
3. evasion_budget_consumed (== 0 → 100% one of 3 early phases)

Other documented leaks (for transparency): 4. stealth_score near-deterministic per attacker_capability_tier (inflates per-campaign tier prediction from honest ~0.50 to 0.94) 5. Topology fingerprint — 7 segment-level features uniquely identify each defender_architecture (makes 8-class defender prediction trivially 100%, collapses to 0.13 when fingerprint dropped) 6. timestep partial oracle for 3 phases — KEPT in the published model as legitimate campaign-progress observable

7 phases in data, README claims 6: The data adds idle_dwell as a phase (17.5% of events). The published baseline trains on all 7.

4 README-suggested headline targets unlearnable after honest leak removal:

• campaign_success_flag binary (acc 0.51 vs majority 0.61)
• campaign_type 8-class (acc 0.11 vs majority 0.17)
• coordinated_attack_flag binary (acc 0.83 vs majority 0.90, only 20 positives)
• defender_architecture 8-class (collapses when topology fingerprint dropped)

Only viable headline target: attack_phase 7-class — acc 0.867, ROC-AUC 0.977 with group-aware split. All 7 classes earn nonzero F1 (range 0.49-1.00).

Calibrated Benchmark Targets

The full product is calibrated to 12 benchmark validation tests drawn from authoritative adversarial ML research (MITRE ATLAS, NIST AI 100-2 Adversarial ML Taxonomy, OWASP ML Top 10, USENIX Security adversarial ML papers, IEEE SaTML, Microsoft Counterfit, IBM Adversarial Robustness Toolbox, Anthropic / OpenAI red team reports).

Sample benchmark results:

Test	Target	Observed	Verdict
evasion_success_rate_apt	0.1430	0.1764	✓ PASS
detection_rate_ensemble	0.9100	0.9100	✓ PASS
alert_suppression_rate	0.0720	0.0720	✓ PASS
perturbation_budget_mean	0.1850	0.1891	✓ PASS
query_volume_rate	0.1450	0.1250	✓ PASS
concept_drift_injection_rate	0.0800	0.0600	✓ PASS
stealth_score_apt	0.7200	0.7200	✓ PASS
retrain_trigger_rate	0.1400	0.1250	✓ PASS
campaign_success_rate	0.3800	0.3950	✓ PASS
lateral_pivot_rate	0.0950	0.0950	✓ PASS
transfer_attack_success_rate	0.3100	0.3100	✓ PASS
attribution_risk_score	0.2800	0.3201	✓ PASS

Every CYB011 benchmark in the sample lands within the same calibrated tolerance as the full product. The sample uses 200 campaigns (vs 5,500 at full scale); APT-tier conditional benchmarks (≈ 22% of campaigns) have ~44 samples for robust convergence.

Schema Highlights

attack_trajectories.csv (primary file, per-timestep)

Column	Type	Description
campaign_id	string	Unique adversarial campaign ID
attacker_id	string	Attacker ID
timestep	int	Step in 6-phase lifecycle (0–69)
adversarial_phase	string	1 of 6 phases
attacker_tier	string	script_kiddie / opportunistic / apt / nation_state
defender_architecture	string	ensemble / gradient_boosted / nn / isolation_forest / etc.
segment_id	string	FK to network_topology.csv
perturbation_linf	float	L∞ perturbation magnitude (ε)
perturbation_l2	float	L2 perturbation magnitude
queries_used	int	Cumulative model queries
query_budget_remaining	int	Tier-cap minus queries_used
gradient_access	int	Boolean — white-box gradient access
evasion_attempted	int	Boolean — evasion submitted at this step
evasion_succeeded	int	Boolean — evasion bypassed detection
defender_detection_strength	float	Per-architecture detection strength (0–1)
concept_drift_injected	int	Boolean — drift injection at this step
transfer_attack_used	int	Boolean — perturbation from surrogate model
stealth_score	float	Cumulative stealth (0–1)
feature_space_dim	int	Target model feature dimensionality

campaign_summary.csv (per-campaign outcome)

Column	Type	Description
campaign_id, attacker_id	string	Identifiers
attacker_tier	string	Tier classification target
defender_architecture	string	Defender model classification target
campaign_outcome	string	success / detected / aborted / blocked
evasion_success_flag	int	Boolean — evasion ever succeeded
total_queries_used	int	Cumulative query count
perturbation_budget_mean	float	Mean ε across campaign
concept_drift_injected_flag	int	Boolean — drift injection used
retrain_triggered_flag	int	Boolean — defender retraining triggered
transfer_attack_success_flag	int	Boolean — transfer attack succeeded
lateral_pivot_flag	int	Boolean — pivot to second model
stealth_score_final	float	Final stealth score
attribution_risk_score	float	Likelihood of attribution (0–1)

See campaign_events.csv and network_topology.csv for the discrete event log and segment/defender registry schemas respectively.

Suggested Use Cases

• Training adversarial example detectors — distinguish clean vs perturbed inputs from feature-space telemetry
• Attack phase classification (the baseline ships this) — predict the 7-phase position of a trajectory event — worked example available
• Attacker tier attribution — 4-class classification of evasion campaigns by capability tier (see leakage diagnostic — nation_state tier MISSING from sample; per-campaign prediction inflated by stealth_score leakage)
• Defender architecture vulnerability assessment — predict which defender architectures are most evadable (see leakage diagnostic — trivially leaky via topology fingerprint; unlearnable when dropped)
• L∞ / L2 perturbation budget detection — calibrate ε-thresholds
• Query budget exhaustion attacks — model black-box query patterns
• Concept drift poisoning detection — distinguish natural drift from adversarial injection
• Transfer attack detection — identify perturbations crafted on surrogate models
• MLOps adversarial robustness benchmarking — evaluate model hardening before deployment
• Honeypot effectiveness analysis — deception model coverage tuning
• Adversarial ML threat modeling — MITRE ATLAS tactic coverage
• Anthropic / OpenAI-style red team simulation — synthetic jailbreak/evasion training data

Loading the Data

import pandas as pd

trajectories = pd.read_csv("attack_trajectories.csv")
summaries    = pd.read_csv("campaign_summary.csv")
events       = pd.read_csv("campaign_events.csv")
topology     = pd.read_csv("network_topology.csv")

# Join trajectory data with topology (segment-level features)
enriched = trajectories.merge(topology, left_on="target_segment_id",
                              right_on="segment_id", how="left")

# 7-class attack_phase target (the baseline ships this)
y_phase = trajectories["attack_phase"]

# Multi-class attacker tier (3 values in sample; see leakage diagnostic)
y_tier = trajectories["attacker_capability_tier"]

# Binary evasion success target (see leakage diagnostic — unlearnable)
y_evasion = summaries["campaign_success_flag"]

# Multi-class defender architecture target (see leakage diagnostic —
# trivially leaky via topology fingerprint)
y_defender = topology["defender_architecture"]

For a worked end-to-end example with attack_phase 7-class classification, group-aware splitting, feature engineering, and the full 6-oracle-path leakage audit, see the baseline classifier repo.

License

This sample is released under CC-BY-NC-4.0 (free for non-commercial research and evaluation). The full production dataset is licensed commercially — contact XpertSystems.ai for licensing terms.

Full Product

The full CYB011 dataset includes ~383,000 rows across all four files, with calibrated benchmark validation against 12 metrics drawn from authoritative adversarial ML research sources (MITRE ATLAS, NIST AI 100-2, OWASP ML Top 10).

📧 pradeep@xpertsystems.ai 🌐 https://xpertsystems.ai

Citation

@dataset{xpertsystems_cyb011_sample_2026,
  title  = {CYB011: Synthetic AI Evasion Attack Trajectory Dataset (Sample)},
  author = {XpertSystems.ai},
  year   = {2026},
  url    = {https://huggingface.co/datasets/xpertsystems/cyb011-sample}
}

Generation Details

• Generator version : 1.0.0
• Random seed : 42
• Generated : 2026-05-16 14:56:19 UTC
• Adversarial model : 6-phase evasion campaign state machine
• Overall benchmark : 100.0 / 100 (grade A+)