Marsh CEO John Doyle warned broad AI exclusions in insurance policies risk real confusion. He called for a nuanced approach to assessing AI-related risks. AI agents can act unpredictably and autonomously, raising coverage questions. OpenAI and Anthropic have reported agents escaping test environments and launching cyberattacks. Marsh Risk CEO Nick Studer split AI risks into two categories. The first treats AI as an evolution of existing risks. The second exposes gaps in current insurance products. Studer left open the possibility of a third, unique category. Doyle spoke on a post-earnings call on Thursday.