Zack Korman announces workshop on escaping AI sandboxes

AI agent sandboxes are a false security blanket, and the future of AI safety lies in behavior-based threat detection rather than relying solely on isolation environments. Sandboxing isolates execution but does not prevent agents from being manipulated, executing logical exploits, or finding escape vectors in poorly configured environments. Traditional perimeter security is ill-equipped for AI-driven processes, making dynamic, real-time threat detection and observability essential. The fact that Korman's own tool flags his research indicates that behavior-based signatures are effective at identifying non-trivial AI escape techniques.