VentureBeat/Gravitee Survey: 88% of Enterprises Reported AI Agent Security Incidents — Only 21% Have Runtime Visibility Into Agent Actions
VentureBeat·high signal
Gravitee's State of AI Agent Security 2026 survey (919 executives/practitioners) reveals a massive enforcement gap: 82% of executives claim policies protect against unauthorized agent actions, yet 88% already experienced incidents. Healthcare is worst at 92.7%. The critical gap is runtime observability — only 21% can see what agents are actually doing in production. Actionable for builders: instrument agent actions with structured logging before deploying to production.