AI Agents: Expanding Enterprise Risk & Data Exposure

Artificial intelligence agents can significantly expand enterprise risk because they have access to data, networks, and cloud services, operating at machine speed. This combination can sharply increase the potential "blast radius" for organizations. Varonis' Matt Radolec warns that enterprises often underestimate what these AI agents can access and how quickly sensitive data could be exposed. He notes that many AI agents can access too much information, and the AI will simply use it without human-like discretion. AWS' Matt Girdharry emphasizes the need to secure not just the AI model environment, but also the data, network, API security layers, and identity and access management policies. What's interesting is that both Girdharry and Radolec discussed how principles like least privilege and runtime guardrails can help reduce the impact of AI agents. They also talked about lessons learned from incidents involving OpenAI and Hugging Face. The bottom line is that as AI agents become more prevalent, securing the entire ecosystem they interact with is crucial for protecting sensitive information.