Insight on Execution-Level AI Governance.

October 16, 2026 · Lois Tullo

Agentic AI has moved the risk: Into the code

Agentic artificial intelligence changes the risk problem for financial institutions because increasingly autonomous systems can move beyond generating an answer or recommendation and carry a decision into action. They can interpret an objective, choose tools, move between systems, adapt when an expected route fails and execute at machine speed. This paper argues that, as those capabilities expand, part of the institution's risk moves into the execution path itself. It develops the Agentic Governance Triangle as a practical way to frame the resulting trade-off among Agent Autonomy, Speed and Scale, and Provable Control and Accountability. Drawing on recent supervisory guidance, emerging runtimegovernance approaches and observed agent behaviour, the paper shows why controls designed around policy, pre-deployment testing and after-the-fact assurance may not be sufficient when an agent can select its own route to an outcome. The Continuous Execution Governance Stack (CEGS) is presented as a runtime governance architecture that carries risk appetite, delegated authority, operating limits, evidence and intervention into the action boundary, the point where an AI decision is about to create an external consequence. CEGS does not remove the trade-offs represented by the triangle. Its purpose is to make them more precise and governable, allowing institutions to expand a safe operating footprint while keeping consequential actions within authority, risk appetite and regulatory expectations. The paper closes with implications for testing, board oversight and the governance of autonomous execution in financial services.