AI agents unlock incredible new efficiencies, but they may also introduce new risks. Google is building AI agents that are useful and secure by design, acting force multipliers, and amplifying human expertise to solve complex problems faster. This talk explores this evolution of artificial intelligence - from simple chatbots to sophisticated AI agents that operate at the intersection of our physical and digital worlds. It outlines Google's principled approach to AI Agent Security, recognizing the dynamic and evolving nature of this critical field and highlighting the critical importance of security, trust, and collaboration.