Agent security incidents reach 54% in enterprise survey

AI agents caused a confirmed security incident or near-miss at 54% of 107 surveyed enterprises, while 69% reported credential sharing somewhere in their agent fleets, according to VentureBeat Pulse Research’s June 2026 survey. The self-selected, mid-market-heavy sample is directional rather than representative, but its 23-point incident-rate gap—63.5% where credentials were shared versus 40.9% where every agent had a scoped identity—gives operators a number to place beside the case for policy inside the agent harness. Builders should treat per-agent identities and isolation as deployment prerequisites, not controls to retrofit after an agent reaches production.