Ben Parker, CEO at eflow
Nearly three in ten firms don’t have a formal strategy for using AI in trade surveillance – or don’t plan to use it at all. Yet seven in ten of those same firms say the AI boom will drive new compliance risks in the next year. This disconnect reveals the real challenge. Firms recognise AI is reshaping how misconduct emerges, but most are building their response as they go, without the governance frameworks or operational maturity to deploy it safely.
The stakes are rising quickly. AI-driven trading creates manipulation patterns that fall outside traditional detection methods, like adaptive strategies, cross-market schemes and behaviour that looks legitimate in isolation. Meanwhile, regulators are making their expectations clear – use AI to identify surveillance risks more efficiently, but be prepared to explain every conclusion it reaches. Accountability sits with firms and individuals, not the technology.
Against this backdrop, the AI boom isn’t just creating opportunities, it’s exposing three critical compliance vulnerabilities that most firms haven’t addressed.
- The detection gap
AI empowers traders with the ability to analyse vast amounts of data and automatically execute trades at incredibly fast speeds. AI trading models, for example, can scan news reports, social media posts and market information to gauge sentiment analysis and predict price movements. Crucially, the use of AI for high-frequency trading moves beyond static rules-based systems and enables firms to execute trades across different markets and assets in milliseconds.
These are exactly the areas where traditional surveillance falls flat. Without using AI-enhanced trade surveillance tools, compliance teams will struggle to spot subtle cross-market manipulation, as traditional systems often focus on specific asset classes and trading venues in isolation. These rules-based systems also work on a predefined understanding of what constitutes abuse and so aren’t able to spot newer forms of abuse that emerge from AI-driven trading habits.
However, using AI for surveillance isn’t a silver bullet. What data AI has access to is just as important. Combining trading data with comms data, for instance, is now a vital way of both detecting and understanding the reasons behind abuse, and it has become a regulatory expectation. What’s more, with the potential for traders to automate trades using AI agents, spotting abuse will only grow in complexity and scale, so systems and data sources will need to evolve to keep pace.
- The explainability problem
The need for AI in trade surveillance has become increasingly apparent. But its use can also create a compliance headache. While it can significantly improve market abuse detection rates, it’s not enough to say that you are now compliant because you have an AI-powered surveillance system in place. Regulators expect transparency and for firms to be able to explain why alerts have been flagged (or not) and the rationale behind these decisions. And there are two main parts to this.
The first part is to understand and explain how a system is calibrated, outlining how it generates alerts and why its alert parameters are appropriate. This comes from evidencing the results of testing and refining the system to show why specific calibrations were decided upon – this is the ‘explainability’ regulators are looking for.
The second part is to provide an audit trail that can illustrate how and why the system made a decision around whether to trigger an alert. This can be a particular pain point for firms. Many ‘AI-powered’ trade surveillance systems on the market use new technologies to make these decisions, but the explanations behind them are unclear and inaccessible to compliance teams, creating an ‘AI black box’ model.
The issue is regulators can’t assess or challenge these decisions, and the lack of explanation can mean investigations don’t progress. Therefore, to overcome this issue, firms must deploy explainable AI systems that enable them to track an alert to its source data and, in doing so, provide context around how the AI arrived at its conclusion.
- The accountability paradox
There is an increased expectation by regulators for firms to use AI – but that doesn’t make the technology culpable for any errors it makes. If AI wrongfully closes an alert without further investigation, for example, accountability for that decision still resides with firms or individuals. That’s why explainability, transparency and human oversight are critical for the effective use of AI-powered surveillance.
While AI can triage alerts efficiently and manage the rising complexity of market abuse and trading activity, it’s not a blanket guarantee of accuracy – as we have experienced with generative AI models, the technology has a tendency to hallucinate. As such, supervisory human oversight is vital for accurate and safe alert clearing, where humans review the system’s outputs and continuously monitor it for quality assurance.
With this outlook, human expertise and judgement have become even more important, not less, especially as the prospect of autonomous AI agents grows as well. If teams can’t explain how they reached a judgement over an investigation, they will be at risk of enforcement action, regardless of whether any wrongdoing has occurred.
Turning AI from a risk to an asset
Recent research into market abuse trends revealed that only 16% of firms had fully deployed AI as part of their trade surveillance strategy. Given the boom in AI adoption for trading, this gap could open up notable risks for firms in detecting market abuse across markets and products and maintaining compliance. But the use of AI for trade surveillance also comes with its own risks. It requires explainable and well-calibrated systems that can accurately demonstrate why alerts were triggered, while working alongside compliance teams and ‘human-in-the-loop’ processes that remain critically important when assessing and closing alerts.
As the use of AI in trading booms, what matters most is how its deployment and use across trade surveillance keeps pace. Tackling these three compliance risks rests on firms balancing AI efficiency with explainability and control. Do this, and they can build a confident and thriving compliance function.