New report finds 98% of security leaders are deliberately constraining agentic AI deployments due to insufficient access controls and identity management.