Artificial intelligence ( AI ) has been gaining traction on trading floors, but concerns over reliability, particularly the risk of hallucination, have remained a major barrier to adoption.

On the trading floor, where decisions are made in milliseconds, AI hallucination is more than a technical flaw; it is a market risk – a single fabricated insight can lead directly to mispriced trades or unintended exposure.

Unlike traditional system errors, hallucinations arrive in fluent, confident language, making them harder to detect. In fast-moving, interconnected markets, that false signal could potentially spread quickly, turning bad data into bad trades at scale. It can also undermine compliance frameworks and expose firms to regulatory and reputational risk.

Against this backdrop, specialist trading analytics firm ExeQution Analytics has launched Eolas, an AI assistant designed specifically for trading, quant and IT teams.

The system has been engineered to be “hallucination-resistant”, ensuring that all outputs are drawn directly from a firm’s approved market and proprietary datasets.

Unlike general-purpose AI tools, which generate responses using probability, the AI assistant converts natural-language queries into API ( application programming interface ) calls. This restricts the system to performing defined functions on authorized data, creating an auditable link between input and output.

The approach is intended to eliminate the possibility of fabricated responses while maintaining the speed and usability associated with newer AI systems.

“Eolas essentially removes all the bottlenecks firms grapple with around data,” says ExeQution Analytics founder and chief executive Cat Turley. “It allows anyone involved in the trading function to speak directly to the data source, in natural language, and for the data to speak back.”

Streamlined workflows

The launch reflects a broader push within financial institutions to extract more value from increasingly large and complex datasets. Conventionally, accessing this data has required specialist support from quantitative or IT teams, often creating delays in fast-moving market conditions.

By enabling direct interaction with both real-time and historical data, the hallucination confining tool is positioned to streamline workflows and improve decision-making speed.

Use cases extend across the trading function. Traders can interrogate execution data intraday, sales teams can generate client reports more quickly, and senior managers can monitor performance while drilling into specific issues as they arise.

The system also integrates with existing entitlement structures, ensuring that data access remains controlled and aligned with internal governance requirements.

For technical teams, the shift could reduce the volume of ad hoc data requests, allowing resources to be redirected towards higher-value work. At the same time, the platform’s design supports compliance and risk functions by providing direct, traceable access to the underlying data used in decision-making.

“The largest data set and most powerful analytics engine in the world is of limited use if only a few people can access it,” says Turley. “Eolas is the next generation of trading intelligence, a true democratization of data that can match the speed of insight to the speed of markets for the very first time.”