Synthetic CISOs Call AI Agents the Most Underestimated Risk
Akamai Technologies, Wednesday, November 4th, 2026
An Akamai survey of 1,067 AI personas modeled on CISOs ranks AI agents last among current concerns but first among underestimated risks.
Akamai commissioned Evidenza to survey 1,067 synthetic CISOs, AI personas calibrated on data about real enterprise security leaders, and checked answers against its own research with security professionals.
Asked to rank 11 current AI security challenges, the personas put agent access to tools and data, including via MCP, last, behind chatbot data leaks and shadow AI.
Yet over-privileged nonhuman identities and agent-driven fraud topped their answers on what the industry underestimates, while 29% named rogue agents as the most disruptive AI attack by 2030, and only 10% reported mature controls for agents versus 48% for APIs.
Akamai recommends inventorying agents, enforcing least privilege and monitoring the APIs they call.