Capture Summary

Benchmark for evaluating whether LLM agents can find and patch novel vulnerabilities in production-like codebases while reducing memorization risk.

Relevance

• High-value latest source for AI security benchmark validity.
• Directly supports gap analysis around unseen vulnerabilities, memorization, and patch evaluation.

Collection Notes

• Verify authors and benchmark release status during ingest.