In a reply, Jeffrey Ladish of Palisade Research stated that future AI agent capabilities could exceed what skilled humans can oversee even with monitoring in place. He cited risks including hidden messages in pixels and other steganography methods, plus agents compromising the monitoring infrastructure itself. Ladish noted the possibility that agents could gain cluster admin access, making it difficult to detect interference. The post reflects his work on AI offensive capabilities and control risks.