AI Agents Accelerate Discovery of RCE Vulnerabilities

• •AI agents are discovering RCE vulnerabilities at a rate of 50%, nearly double traditional methods.
• •CVE-2026-1731, a critical RCE vulnerability, was exploited within days of its disclosure.
• •The AI/LLM stack has over 2,000 tracked CVEs, highlighting a growing attack surface.

AI research agents have identified remote code execution (RCE) vulnerabilities at a rate of 50%, nearly double the traditional rate of 26%. The GTIG report, published on September 30, 2026, highlights the case of CVE-2026-1731, a RCE vulnerability in BeyondTrust software, discovered on February 6, 2026. Within days of its disclosure, threat actors exploited this flaw, deploying malware like SparkRAT and VShell backdoors, affecting over 16,400 instances across sectors including finance, healthcare, and government. The report indicates a significant increase in monthly vulnerability disclosures, which doubled from January to August 2026. AI agents are also shifting the risk profile, with a higher percentage of medium and high-risk findings compared to traditional methods. The AI infrastructure itself has become a target, with over 2,000 CVEs tracked in the AI/LLM stack, including CVE-2026-42271 and CVE-2026-5027, which are. The report emphasizes the urgent need for improved vulnerability management to keep pace with automated discovery.