Banks Tighten Governance as Agentic AI Workflows Expand

Banks are moving beyond AI models built mainly for scoring, prediction and classification toward agentic systems that can plan tasks, invoke tools and act across multiple platforms with limited supervision. The technology could accelerate lending, fraud detection, compliance and customer service, but its autonomy raises the risk of cascading errors, excessive access and unclear ownership. Regulators and risk teams therefore need controls that identify each agent, preserve an end-to-end audit trail and assign human accountability for consequential actions.

Deloitte Center for Financial Services said on March 5, 2026, that one in three financial institutions was allocating budgets to agentic AI. Wells Fargo is working with Google Agentspace, while PNC Financial is using orchestrated agents in developing a mobile app; Goldman Sachs, JPMorgan Chase, Citi and BNY are also investing in the technology. Deloitte identified more than 350 risks linked to autonomous behavior in the MIT AI Risk Database and recommended unique agent IDs, immutable tool-use logs, least-privilege access, real-time monitoring, escalation triggers and designated owners, validators and stewards.