JSON (full dataset) CSV (one row per risk)

License and attribution

The dataset is licensed Creative Commons Attribution-ShareAlike 4.0, because its framework mappings are adapted from OWASP material under that license. If you reuse it, credit "AI Security Risks (aisecurityrisks.com)" and OWASP as described in the file's attribution field, and share your adaptation under the same license.

What's in the JSON

risks: slug, title, summary, OWASP LLM 2026 ID, OWASP Agentic IDs, ATLAS technique IDs, NIST AI 600-1 categories, the uses each risk applies to, controls with their source, vendor questions, and the cases linked to it.
cases: title, disclosure date, ATLAS case-study ID, incident or research demo, summary, outcome, and source IDs.
sources: title, publisher, URL, type, publication and retrieval dates, and how much of it we read.

The data changes when a framework or ATLAS release changes. Each file carries the framework versions it was built against. The cases are also in the RSS feed.