According to CoinMeta, as reported by Forkast, AI agents detected remote code execution (RCE) vulnerabilities at a rate of 50%, which is nearly twice the 26% rate in the broader CVE ecosystem. This finding is detailed in the GTIG report released on September 30, 2026, and highlights the shift in how software defects are discovered and subsequently weaponized. The report indicates that AI-powered autonomous research agents excel in complex semantic code path reasoning and dynamic proof tool synthesis, enabling them to detect memory corruption and logical bypasses in core libraries and at runtime.