I talk to a lot of startups building security tools for AI agents, and my advice to all of them is, treat AI agents like malware. Even if agents provide a lot of value to those who use them, agents are a security nightmare to those who have to secure them.
