ServiceNow patches critical AI platform flaw that could allow user impersonation

ServiceNow patched a critical vulnerability (CVE-2025-12420, CVSS 9.3) affecting Now Assist AI Agents and the Virtual Agent API that could allow unauthenticated user impersonation and unauthorized actions; AppOmni discovered the flaw and fixes were deployed to most hosted instances on Oct 30, 2025 with patches provided to partners and self-hosted customers. The accompanying research also revealed that default agent-discovery settings can enable second-order prompt-injection attacks where low-privileged data fields are later processed by higher-privileged AI agents, enabling data access/modification and potential privilege escalation; ServiceNow has updated documentation and recommends mitigations such as agent segmentation, human supervision for powerful agents, and monitoring agent behavior.