By Kirat Singh, President of Risk and Alternative Assets, Clearwater Analytics

On the face of it, AI looks like it fits the derivatives industry like a custom-made glove. After all, OTC derivatives markets are, regardless of the asset class in question, built on data, models and calculations, which is exactly where AI is set to revolutionise how efficiently financial institutions can get their work done.

Understanding how risks interact takes more than calculating a specific risk, and complex derivatives expose one of the fundamental challenges of bringing agentic AI into financial markets. An OTC derivative position rarely contains a single exposure. An interest rate swap, as a prime case in point, has numerous sensitivities across different points of the curve. In FX, a bilateral derivatives portfolio can bring volatility and potentially correlation risk alongside the direction of the underlying market. In commodities, derivatives can introduce physical exposures. The point is that regardless of the type of portfolio in question, firms have to consider counterparty. liquidity, collateral, margin and funding risks, which interact with each other and can change drastically as markets move.

Asking an AI agent what your risk might be becomes much more complicated. Sure, all AI agents will give an answer, but what a risk manager cares about is how it got to that answer. It is the equivalent of showing your working to a maths teacher in school. The questions range from what market data was used and which valuation models were called, to whether assumptions were consistent between products and asset classes. And that is all before asking whether the agent has captured basis and liquidity risks, and what happens to collateral and margin requirements under stress.

For those immersed in OTC derivatives markets, models and assumptions have always been central to how swaps and other bespoke instruments are priced and managed. Agentic AI makes model discipline even more important, because it allows analysis to be produced at extraordinary speed. That is why the most promising application of AI in derivatives is to let agents orchestrate the trusted models, data and calculations that firms already use.

Take the valuation of a complex instrument, for example.  The AI Agent should receive an approved valuation model, defined market data and explicit calibration parameters, which fix how the instrument is priced. Only then can the agent interrogate the findings and combine them with other portfolio information. The intelligence can be flexible while the financial mathematics remains controlled.

This becomes particularly powerful as financial infrastructure grows more API-driven. Positions, reference data, market data, valuation engines and risk models can be made available as defined services. An agent can yoke them together, letting risk managers question exposures without switching between multiple systems and reports. They can then spend more time challenging the information that the agent has gathered for them.

The history of derivatives markets (who could forget 2008?) reminds us that losses often emerge not from top-line risk figures, but instead from leverage, liquidity, basis, correlation or funding assumptions buried elsewhere in the portfolio.

Agentic AI could become an extraordinarily powerful tool for finding those connections. But in the complex world of derivatives, knowing the individual risks matters less than understanding the intricate relationships between them.