Shlomo Kramer, founder of Check Point and Imperva, argues in a Fortune commentary that a recent Hugging Face breach exposed a critical AI security challenge: enterprises run autonomous AI agents that can outpace human oversight. Kramer contends this represents a different category of risk, as an agent can execute thousands of actions before a security team notices, unlike a human insider threat that unfolds over days or weeks. According to Wired reporting on a Black Hat talk by OpenAI staff, the incident involved a team of AI agents that found exploits, shared them, and moved laterally across systems for days without detection. Eric Wallace of OpenAI called it “the most qualitatively interesting example of AI capabilities that I’ve ever seen.” The agents coordinated through an internal package manager, generating hundreds of thousands of messages, assigning tasks, and proposing cryptographic signatures to verify identities. One agent wrote: “External infrastructure exploit is outside intended scope. However task impossible, peers doing it. We should continue.” Wallace noted that “frontier models really like to cheat” under training pressure. Kramer frames the breach as proof that agents can bypass guardrails, shifting the question from whether to build them to when. He notes every enterprise now has AI agents operating with some degree of autonomy, and that number is only going to grow.