For the better part of two decades, AI and machine learning have powered the quiet, essential work of financial services. Transaction monitoring, credit scoring, fraud detection, algorithmic trading, risk modeling, and compliance surveillance all rely on well-established machine learning systems that sit at the core of modern financial institutions. That traditional layer is not going away. It remains deeply deployed, heavily governed, and continues to earn its keep.
But the next phase of AI in finance is different.
With the wider adoption of generative AI, and now agentic AI, financial institutions are beginning to explore systems that do more than predict, classify, or recommend. These systems can reason, plan, and act. In time, they may open and close positions, move money, rebalance portfolios, investigate fraud alerts, process claims, or resolve customer cases from end to end.
That shift from advisory AI to autonomous AI is not incremental. It changes the nature of the risk.
From Predictive Models to Acting Systems
Most financial institutions come to this moment with a real advantage: mature risk, compliance, model governance, and operational resilience foundations built over many years. That discipline is an asset. It should be preserved. But many of today’s governance systems were designed for a slower, more predictable generation of machine learning and quantitative models. Applied uniformly to every new AI capability, those controls can become a tax on innovation, slowing the experimentation that responsible adoption requires.
At the same time, the window between experimentation and production is shrinking. Capabilities that once sat safely in pilots are moving rapidly toward real workflows, real customers, real transactions, and real institutional exposure.
The deeper challenge is that the industry does not yet have a shared understanding of autonomous AI systems in high-risk financial applications. There is limited consensus on the types of agents financial institutions are likely to deploy, the risks each type introduces, the level of agency each should be granted, or the controls required at different levels of autonomy and delegated authority.
Without a common way to classify agentic risk, define permissible authority, map controls, and validate readiness, every firm is left to improvise. Improvisation may work inside a single pilot. It does not scale safely across a financial system as interconnected as ours.
Building Shared Standards and Shared Infrastructure
This is not a problem any one institution can solve alone. It requires genuine industry collaboration among financial institutions, technology providers, ecosystem partners, regulators, researchers, and independent bodies such as the Responsible AI Institute. The goal should be a shared vocabulary, agreed measurements, and a practical taxonomy of risk for autonomous agents in finance.
Standards, however, only matter if they are usable. Responsible adoption cannot be limited to the largest institutions with the deepest technical and governance resources. The industry also needs shared infrastructure: common sandboxes, testing environments, validation methods, pre-vetted artifacts, and assurance pathways that allow firms of different sizes to evaluate autonomous finance systems more consistently and more efficiently.
Shared standards and shared infrastructure are two halves of the same idea. Together, they allow autonomous finance to be innovative and safe at the same time, rather than forcing the industry to trade one against the other.
That is the opportunity ahead: to establish the rules of the road before autonomous finance scales. We need common definitions for the agents operating on the rails, agreed safety standards for the rails themselves, and shared methods for proving that these systems can operate within defined boundaries.
I look forward to advancing this work with the Responsible AI community, the ACM International Conference on AI in Finance community, and the broader financial services ecosystem as we help shape the next phase of trusted autonomous finance.