Earlier this year, research revealed that only 16% of senior regulatory professionals said they had fully deployed AI as part of their trade surveillance strategy. There was then a divide between those actively rolling it out in specific functions (31%), those planning deployment within the next one to two years (24%), and those yet to form a formal strategy or plan for its use (29%).
The impact and deployment of AI is moving so fast that these stats will likely be changing on a month-to-month basis and could already be different from when the survey was conducted in January 2026. But what they do show is the gap many firms have between moving from AI as a theoretical idea to a tangible reality in their trade surveillance function – and many compliance teams will still be on this journey.
Over the last couple of years, however, the increased sophistication of AI means it is now a tool firms are expected – by regulators and fellow industry professionals – to have a strategy in place for. That’s because, through its targeted and measurable use, AI has been able to address some of the most common pain points firms have faced, such as the ‘noise’ generated by high alert volumes and limited insights into the behaviour that accompanies suspicious trades.
Crucially, by judging it on practical outcomes, firms can overcome the perception held by some that AI is little more than ‘innovative hype’. Yet this perception can only be tackled when AI is successfully integrated into workflows and has the necessary transparency, control and human guardrails in place. These are features that enable it to enhance a compliance team’s work rather than become another risk in its own right.
How AI enhances trade surveillance operations
Understanding how AI can improve trade surveillance is key to successfully integrating it into workflows. One of its key benefits is to help focus analysts’ time on the tasks that matter most. For example, the volume of false positive alerts that analysts have to deal with on a day-to-day basis has traditionally been a huge resource drain, particularly for smaller teams. However, by using AI to supplement rules-based workflows, they can manage the triage process more efficiently, focus on genuine risk instead of slow and repetitive tasks and make their work more effective.
With this approach, processes become more streamlined, investigations take place faster and decision-making becomes more consistent. Analysts can then use AI to gather additional insights faster. For example, it can be used to interrogate data and uncover hidden context to support decision-making. But for all of these benefits to be realised, AI needs to be built into established processes.
Embedding AI into existing workflows
One of the challenges of using AI in trade surveillance has been its lack of integration into existing systems. People have been chopping and changing between different AI models to get insights, but AI ideally needs to be integrated within trade surveillance infrastructure to make the process as frictionless as possible – it has to be built in instead of layered on top. This also means it should be aligned to workflows to support how surveillance teams already operate, and it needs to be controlled by design, so that it’s governed, auditable and consistent.
Critically, AI will only work well when applied to a strong data foundation and interoperability between systems, so it can easily access any context behind an alert. This means an alert’s lifecycle has to be available in the audit history, creating clear visibility into decisions and enabling teams to review and challenge outcomes at every stage. With AI embedded in this way, the workflow should run as follows: data is inputted into the system, AI processes this data and prioritises it, the analyst reviews this data and then this is added to the compounding audit trail, which can be reviewed at any time.
Human-in-the-loop principles and working with AI
The orchestration of AI use in trade surveillance so far has really highlighted the importance of the principle of ‘human-in-the-loop’ decision making. It’s crucial firms understand that AI’s not there to autonomously make decisions, but to improve how human judgement is applied. While it can do a lot of the repetitive work, this needs to take place under human supervision – AI is still prone to errors and hallucinations, making human oversight crucial for quality assurance, and it definitely shouldn’t be used to apply judgement to cases. Instead, the most effective use of AI is when it’s used to support human processes.
For example, analysts could open up an AI chat interface in their trade surveillance system, similar to generative AI models on the market. They could then input a prompt like “list five traders with the most failed alerts”. AI will traverse the data and find the information that best answers the question. It’s also possible for AI to look across a broad range of different conditions, like trader behaviour patterns and cross-asset class investigations. Analysts could also ask AI for useful context like finding out what parameters were breached and whether it was a major or minor breach.
Making the move
Despite the divide in AI strategy, the research also shows that over 70% of the market will have deployed AI in the next year or two. So, the important thing will be how this deployment happens and how firms make the move from theory to reality. What’s crucial is that a strict ring fence is applied, so that data doesn’t leave the controlled system, and that the AI model being used is fully auditable and explainable – e.g. it’s possible to see how AI produced an output from the source data. This foundation will be essential for integrating newer AI capabilities.
It’s clear that AI will become more deeply embedded across surveillance and related controls, and there will be a stronger focus on measurable outcomes. This will be joined by a need to meet regulatory expectations – firms will need to clearly explain how AI supports decisions, with consistent governance and auditability. This will be particularly important as the next phase of AI in trade surveillance takes place, which will see greater alert explanations, newer capabilities such as batch alert closure, and AI being applied to comms surveillance.
AI adoption is accelerating, and its use is key to overcoming some of trade surveillance’s biggest challenges. But what matters most is that this adoption is controlled and that AI is fully embedded into existing workflows. That will decide how well AI becomes a reality in a firm’s trade surveillance operation.
