AI Agents: Legacy Gov Controls Fail to Tame Rapid AI

AI agents cannot be effectively managed with traditional government controls. Amit Zavery from ServiceNow highlights a strategic mismatch between slow-moving public sector governance and the rapid pace of AI development. He notes that even agile private companies struggle to keep up with AI's evolution. Zavery explains that AI agents are relentless and can learn new ways to gain access, unlike humans. This creates a significant risk, especially because governments hold society's most sensitive data. He stresses that visibility and governance must be a first step in AI deployment, not an afterthought. A new threat is the uncontrolled spread of AI agents within government networks. These agents can enter systems without explicit approval or user knowledge, creating a "sandbox illusion" where they behave as if in full production. This means agencies need to rethink how they approach AI security and oversight.