Analysis · curated 13 Aug 2026

Towards Automated Vulnerability Discovery and Repair with Safety-Governed AI Agents - SPAR Project

Coverage timeline

13 Aug 2026sparai.org

Single-source analysis — first reported, latest, and curated coincide.

Why it matters

The SPAR project targets a core agentic-security concern — governing autonomous agents' tool use and high-risk actions during vulnerability research — which defenders must address as AI agents gain access to real code repositories and execution environments.

The SPAR project 'Towards Automated Vulnerability Discovery and Repair with Safety-Governed AI Agents' is a research initiative to build safe, expert-in-the-loop AI agents that discover, verify, and patch code vulnerabilities. It proposes three components: an automated vulnerability-research agent, a safety harness that governs agent permissions, tool use, and high-risk actions with human approval and audit logging, and a security data engine capturing full expert-agent-tool trajectories, drawing inspiration from Microsoft's MDASH multi-model scanning system.