Commercial AI Models Show Rapid Gains in Vulnerability Research

Forescout Verde Labs tested 50 AI models and found that modern commercial and some open-source models can complete vulnerability research and, for half of tested models, generate working exploits; using RAPTOR and other tools they discovered four new zero-day vulnerabilities in widely deployed OpenNDS. The report warns that AI-driven discovery and exploit generation lowers the barrier for attackers, making it likely organizations contain unknown vulnerabilities that AI will find and potentially weaponize.