Seventy percent of enterprise AI deployments operate outside security governance controls, creating unmanaged non-human identities with excessive permissions, unvetted data egress to third-party LLM APIs, and prompt injection attack surfaces that bypass CASB and DLP controls. This is a structural architecture and governance risk affecting enterprise SaaS platforms, LLM integrations, and agentic AI workflows across all sectors — not a discrete vendor vulnerability. CrowdStrike Falcon AIDR and Falcon Platform are referenced as detection tooling in the source intelligence.