Black Hat USA 2026: Will vulnerability discovery eventually decline in the AI era?

Researchers at Arizona State University have demonstrated the significant impact of AI on vulnerability discovery, using advanced models like Anthropic's Claude Mythos to find hundreds of flaws in the Linux kernel. The team found that AI models, especially when enhanced with workflows and trained on past vulnerabilities, can discover vulnerabilities at a rate that outpaces human reporting capabilities. This rapid discovery raises concerns about responsible disclosure and the ability of organizations to patch systems effectively, potentially leading to increased cybercrime or system instability.