AI Agent Failures Create a New Uninsurable Risk Class

An audit of an autonomous AI agent fleet found 34 user-facing features that had silently stopped working while dashboards stayed green. These non-adversarial, correlated, and silent failures represent a loss class that existing security products and insurance policies are not designed to cover. The author argues that shared model infrastructure and common misconfigurations create systemic risk that traditional risk instruments miss.