Preparing for Agentic AI – the Next Frontier in Risk Management

For corporate treasury, 2025 brought an unprecedented level of uncertainty and volatility. Trade disruption and geopolitical tensions dealt a serious blow to confidence in markets and forecasts, while the exponential growth of artificial intelligence (AI) and cryptocurrencies simultaneously posed both new risks and new development avenues for companies to consider. Perhaps unsurprisingly, many organizations were left paralyzed while attempting to navigate last year’s business environment.

Now, a quarter into 2026, finding concrete ways to prepare for and mitigate the impact of volatility in the external environment remains a priority for most treasury teams. Some are turning to agentic AI, which has the potential to help corporate treasurers optimize risk management.

AI’s Evolution in Finance

In its earliest applications within corporate treasury, AI played a largely assistive role. Systems were designed to support human decision-making by processing large datasets, highlighting anomalies, and improving the speed and accuracy of data analysis. Leveraging assistive AI in cash forecasting, scenario modeling, and exception reporting helped treasury firmly with the treasurer.

Over the past few years, those capabilities have expanded. Machine learning models have begun to identify correlations across markets, stress-test a company’s exposures under multiple possible scenarios, and continuously update forecasts as new information becomes available. These tools delivered more timely insights than earlier iterations of AI, but they still operated within the same basic framework: AI informed people, who made the decisions.

Agentic AI represents a meaningful shift away from this model. Rather than simply analyzing data and offering recommendations, agentic systems are designed to reason, decide, and act within clearly defined boundaries. This can include rebalancing portfolios as liquidity conditions change or flagging emerging risks that require immediate attention. AI agents can also automate and streamline hedging workflows, notably to adjust hedge positions in response to market movements. Particularly when it comes to FX, using agents to run compliance assessments between exposures and hedging policies will offer notable benefits for treasury teams. Agents can suggest hedging derivatives that would bring outstanding exposures in line and can even be equipped to create those trades within a treasury management system.

Real-time Risk: Growing Pressure on Corporate Treasurers

Such capabilities align with companies’ growing expectations of their treasurers. Periodic risk assessment is no longer satisfactory: Treasury teams must now have continuous awareness of potential exposures and faster responses to emerging threats. AI systems that can monitor, reason defined parameters, to help manage risk in real time.

In recent industry research on AI adoption in finance, KPMG found that nearly half of the organizations surveyed are piloting or actively using AI in treasury and risk management operations, with early use cases delivering measurable improvements in forecasting accuracy, anomaly detection, and risk insights.

Progression from assistive tools that support human decision-making to agentic systems that operate with a degree of autonomy is an essential step as risk environments become more dynamic. This is especially evident when we consider that the ability for corporations to respond at speed to new challenges will be not only a competitive advantage, but also fundamental in weathering volatility storms.

Still, we cannot ignore the fact that greater autonomy raises the stakes when it comes to entrusting AI with decision-making. The more responsibility AI is given, the more critical it becomes that AI systems operate on a complete, accurate, and contextual view of risk. In fact, failing to understand how to use AI responsibly creates risk itself. When AI systems move beyond insight generation to decision execution, the tolerance for poor data, fragmented systems, or unclear governance falls sharply.

Before making the move from assistive analytics to agentic decision making, treasurers must ask themselves one important question: Is my organization constrained by inadequate data foundations? These two factors can help treasury teams answer that question ‘no’:

Strong data foundations are not optional; they are the differentiator between agentic systems that enhance risk management and those that introduce new sources of exposure.

In many organizations, financial data remains fragmented: Risk, liquidity, and exposure information is spread across multiple enterprise resource planning (ERP) systems, treasury platforms, and business units, each with its own definitions, update cycles, and controls. While these environments can support traditional, human-led decision-making, they clearly limit what agentic AI can responsibly achieve.

If exposure data is incomplete, delayed, or misaligned, AI models will fill in the gaps—likely introducing inaccuracies. In a risk management context, assumptions can quickly compound. Decisions that may appear to be rational when based on a narrow dataset may be misaligned with the organization’s true risk profile or policy constraints, yet they may be the actions an AI system takes if it lacks a complete and accurate view of the organization’s financial position.

For example, an AI agent might see a large net U.S. dollar (USD) receivable and decide to hedge it with FX forwards, not realizing that unbooked USD payables and intercompany loans will soon offset much of that exposure.

Acting on incomplete data, the agent will overhedge, creating a synthetic short position once the missing exposures materialize. The result is unnecessary P&L volatility and new risk introduced by a decision that appeared rational based on a limited view of the company’s FX situation.

MCPs put AI in context. High-quality data is a prerequisite, but it is not sufficient on its own to assure management that decisions are optimized risk policies, liquidity thresholds, approval hierarchies, and regulatory constraints that shape how the organization manages its financial risk in practice.

This is where model context protocols (MCPs) play an increasingly important role. In simple terms, MCPs provide a structured way for AI models to interpret internal data within the correct organizational and policy framework. Rather than treating data as isolated inputs, MCPs help ensure that AI systems understand what the data represents, how it should be used, and where boundaries apply.

An MCP enables an AI agent by providing the decision context it cannot infer from data alone—codifying risk policy, liquidity limits, approval rules, and regulatory constraints in a form the model can reason over. For an FX hedging decision, this means the agent understands which exposures are hedge eligible, how much can be hedged, and what funding or governance limits apply. With this context, the agent can recommend a policy compliant partial hedge and escalate exceptions, rather than mechanically optimizing based only on exposure data. In effect, the MCP turns the agent from a pattern matcher into a constrained decision Maker aligned with how treasury actually manages risk.

Introducing MCPs is a critical shift from a risk management perspective because they embed governance directly into how agentic AI makes decisions. By encoding policies, limits, and approval rules as decision context, MCPs make AI behavior predictable, explainable, and auditable: Each action can be traced back to the constraints that allowed or blocked it. At the same time, they allow treasury teams to retain control by defining the boundaries within which automation operates, shifting human before adopting agentic AI in treasury operations, corporate treasurers must recognize the need for intelligent and more automated from the outset.

Without these requirements in place, organizations should not attempt to deploy agentic AI at scale.

So, Is Agentic AI the Answer?

The potential for agentic AI to revolutionize risk management is an opportunity that treasury teams should not overlook. Treasurers risk being left behind if they do not invest time and resources now to put these systems in place.

That said, organizations seeing early value from advanced AI capabilities share a common trait: They have invested in connecting systems, standardizing data definitions, and establishing clear governance over critical risk and exposure data. And integrated data and governance frameworks lay the foundations for systems to evolve and develop alongside AI, rather than lagging behind it. As a result, these tools operate on trusted information, allowing insights and actions to be interpreted and audited with confidence up front. Preparedness in data foundations, coupled with contextual protocols, can help ensure that agentic AI works for treasurers—and that it is an asset, rather than an added risk for the organization.