Capture Summary

Meta/Purple Llama CyberSecEval 3 publication extending cybersecurity risk and capability evaluation for LLMs, including offensive cyber capability areas.

Relevance

• Important benchmark family for measuring cyber capability and model risk.
• Useful for comparing evaluation coverage with Cybench, CVE-Bench, and AIxCC.

Collection Notes

• Verify authors and paper date during ingest.
• Ingest should distinguish model-risk findings from AI-for-security benchmark utility.