Solidity REKT Dataset: Adversarial Smart Contract Security Training Data

A collection of 64,691 samples for training specialized AI models in smart contract security. The dataset is structured for three tasks: vulnerability classification, security auditing with explanations and patches, and exploit proof-of-concept generation. It was created by aissacas and last updated on 2026-05-20.

Use Cases

• Train a vulnerability classifier based on labeled smart contract code for type and severity.
• Develop a security auditor model based on data for generating explanations and patches.
• Build an exploit proof-of-concept generator based on adversarial examples.
• Benchmark model performance on specialized smart contract security tasks.

Strengths

• Contains 64,691 total training samples across three specialized splits.
• Provides data for three distinct security tasks: classification (27,616 samples), auditing (34,664 samples), and exploit generation (2,411 samples).
• Sourced from a known audit dataset (oxdev/smart-contract-security-audit-v2) with 48.4K rows.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Derived from oxdev/smart-contract-security-audit-v2, which aggregates audits from platforms like Code4rena and Sherlock.

Collection Method

Likely processed and split for specific model training tasks.

Freshness

Last updated 2026-05-20 15:59:00; freshness should be verified.