Avayam is a research-grade cybersecurity benchmark that evaluates AI agents on their ability to remediate real-world vulnerabilities. It agentifies the MSR 2020 dataset (Fan et al.), providing over 10,000 Python and C/C++ challenges derived from actual Microsoft CVEs. Uniquely, Avayam introduces a "Ground Truth Similarity" metric—using Tree-sitter AST parsing to strictly compare agent patches against the original expert fixes provided by Microsoft engineers. This ensures that agents are scored not just on passing tests, but on adhering to secure coding standards and reproducing canonical security patches