BCG: 73% of Big Banks Run AI Agents in Production, Only 8% Can Monitor Them

Seventy-three percent of the world’s systemically important banks now allow AI agents in production for select use cases, and only 8% have a comprehensive, standardized monitoring and control framework that explicitly covers them. Both numbers come from BCG’s 2026 AI safety and governance benchmark of 26 global and domestic systemically important banks, and the distance between them is the story.

Usage is still narrow. Ninety-two percent of the banks are piloting or permitting agentic AI, but none permits broad production use, and at 88% of them agents make up 10% or less of the AI portfolio. More than half — 54% — rely on restrictions or prohibitions written into their risk-appetite frameworks. Problems are already showing: 46% reported an AI-related issue or incident in the past 12 months, and more than a quarter said AI systems were falling behind their promised business case. Fewer than half keep a centralized, tool-based AI inventory, and none has a consistently implemented human-oversight framework.

BCG’s headline finding is that governance and speed are not a tradeoff. Banks with more mature risk management are 2.5 times more likely to be advanced AI adopters, and banks using pre-approved patterns for lower-risk applications are nearly twice as likely to move them from idea to production in under six months. Coauthor Anne Kleppe says governance now has to become an operating capability.

That framing travels beyond banking. Seven Forrester research directors make the same point from the risk side: deciding how much autonomy to grant an agent is a business decision, and delegating work does not remove human accountability. The practical move for any company scaling agents is the one the faster banks made — an inventory of what is running, a short list of pre-approved patterns for low-risk work, and monitoring that watches what agents actually do, not only policies that forbid them.