akshaynexus/DeFiHackLabs-Dataset

---
language:
- en
license: mit
pretty_name: DeFiHackLabs AI Incident Analysis
tags:
- cybersecurity
- smart-contracts
- defi
- tabular
- text
- datasets
annotations_creators:
- machine-generated
language_creators:
- machine-generated
size_categories:
- n<1K
source_datasets:
- original
task_categories:
- text-classification
- summarization
task_ids:
- multi-class-classification
configs:
- config_name: incidents
data_files:
- split: train
path: output/incidents.jsonl
---

# DeFiHackLabs AI Incident Analysis

This dataset contains incident-level DeFi exploit records with AI-generated analysis fields.

## Data Source

Records are derived from DeFiHackLabs PoC tests and normalized into incident rows.

## Included Split

- `train`: `output/incidents.jsonl`

## Row Selection

Rows in `output/incidents.jsonl` are filtered to keep only complete AI-analysis entries:

- `ai_analysis` exists in source records
- status is `resolved` or `partial`
- non-empty `explanation`, `root_cause`, and `vulnerability_type`
- `attack_steps` is a non-empty array

The exported JSONL flattens analysis fields to top-level columns for better Hub table browsing.

## Main Fields

- `id`
- `title`
- `attack_title`
- `poc_code`
- `resolution_status`
- `resolution_evidence`
- `resolved_at`
- `num_contracts`
- `num_verified_contracts`
- `num_source_contracts`
- `num_bytecode_contracts`
- `ai_explanation`
- `ai_root_cause`
- `ai_vulnerability_type`
- `ai_attack_steps`
- `ai_attack_steps_text`
- `ai_confidence_score`
- `ai_confidence_reasoning`
- `ai_confidence_verified_contracts`
- `ai_confidence_has_source_code`
- `ai_confidence_known_pattern_match`
- `ai_mitigation`
- `contracts`
- `ai_analysis`
- `metadata`

## Intended Use

- exploit pattern modeling
- vulnerability type classification
- incident summarization and root-cause training