0DIN now releases weekly versioned JSONL datasets for AI security research, enabling real-time threat detection and response. The corpus includes structured vulnerability disclosures with metadata, exploit prompts, and detection signatures.

In brief - The 0DIN dataset provides a weekly updated, versioned JSONL corpus of AI security disclosures, including vetted threat intelligence with exploit prompts, responses, and detection signatures to enhance AI model defenses.

Technically - The dataset is split into general disclosures (metadata like vulnerability type, affected model, and risk summaries) and a vetted subset with actionable intelligence: triggering prompts, confirmed exploit responses, sector-specific variations, and detection signatures. Exports include typed schemas, timestamped manifests, and publisher git SHAs for reproducibility, supporting integration into security pipelines against threats like prompt injection (e.g., CWE-1324) and model exploitation.

Source: https://0din.ai/blog/public-disclosures-corpus

#Cybersecurity #ThreatIntel