TL;DR
Goldman Sachs and Deutsche Bank are testing agentic AI systems for trading surveillance, according to Bloomberg. Unlike traditional rule-based monitoring, these AI agents can analyse trading behaviour across multiple signals in real time, compare it with historical patterns, and flag conduct that may need human review.
Beyond static rules
Current bank surveillance systems typically work by matching trades against predefined criteria: if a trade exceeds a certain size or fits a known risk pattern, it triggers an alert. Compliance teams then review manually.
The problem is scale. Modern markets generate enormous volumes of data across asset classes, time zones, and venues. Static rules produce large numbers of false positives, while subtle manipulation may not match any known pattern. Agentic AI systems aim to address both issues by examining trading behaviour across multiple signals and detecting unusual combinations of actions.
How the banks are approaching it
Deutsche Bank is working with Google Cloud on AI agents that monitor trading activity in near real time, reviewing large sets of order and execution data and flagging anomalies. The system looks at relationships between trades, timing, market conditions, and trader history — not single events in isolation.
Goldman Sachs is taking a similar approach, extending its existing AI investment in trading and risk systems into compliance. The bank’s focus is on AI agents that can operate with a degree of independence in scanning for misconduct indicators, identifying patterns that do not fit a clear rule but still stand out as unusual.
In both cases, human compliance staff remain responsible for reviewing flagged cases and determining next steps.
Regulatory context
US and European regulators have encouraged firms to improve monitoring of market abuse and manipulation. While rules do not mandate agentic AI specifically, they require effective systems and controls. If AI tools help meet that standard, adoption is likely to grow — though banks must also ensure models are explainable, free from bias, and able to withstand regulatory scrutiny.
Looking forward
If agentic surveillance proves effective, it could shift how compliance teams operate — spending less time sorting simple alerts and more time evaluating complex cases surfaced by AI. The need for human judgement does not disappear, but where that effort is focused may change significantly.