Enterprise AI agents lack proper governance and security
Companies deployed AI agents without controls, now scrambling to catch up.
1Uncontrolled AI agents
Enterprises knowingly deployed AI agents without governance controls. VentureBeat research shows 57-68% plan to switch or add vendors within 12 months across five control layers (identity, evaluation, cost telemetry, context layer, orchestration). Significant gaps exist in each area.
2Chatbots in agent clothing
71% of companies admit only a quarter of their so-called 'agents' can complete multi-step tasks independently. Just 10% say true agents constitute most deployments. This reveals many firms are rebranding advanced chatbots as 'agents' without meeting key criteria.
3Autonomy outpacing trust
Two-thirds of enterprises already allow or plan to allow AI agents to push code changes to production without human review within a year. Yet only 5% fully trust the evaluations enabling this. Half experienced customer-facing failures from internally approved agents last year.
4Credential sharing risks
69% of organizations permit some agent credential sharing. These companies experienced security incidents at 63.5% rate versus 40.9% for those assigning unique identities. Scoped identities for production-touching agents should be priority one for risk reduction.
5GPU underutilization
Over 80% of companies running GPUs report ≤50% utilization. Only 44% rigorously track AI compute costs/returns. Before buying more hardware, enterprises should optimize existing GPU workloads - the low-hanging fruit for cost efficiency.
6Context governance gap
57% traced incorrect agent answers to ungoverned business data (wrong metrics, stale definitions). Standardizing metrics and entity definitions must precede agent scaling. Investment is expected to shift toward context layer solutions in coming quarters.