OpenAI CEO, Sam Altman, warns of the increasing real-world risks posed by powerful AI agents, including their potential misuse in finding critical security flaws.