Goldman Sachs and Deutsche Bank are testing agentic artificial intelligence systems to monitor trading activity in real time. The shift signals a move beyond static rule-based alerts toward pattern recognition that could tighten oversight in complex markets.
According to Bloomberg, both banks are exploring or deploying AI agents designed to analyze orders and executions as they occur. Traditional surveillance systems typically trigger alerts when trades breach predefined thresholds, leaving compliance teams to manually review flagged cases. Deutsche Bank is working with Google Cloud to build agents capable of scanning large volumes of order data and identifying anomalies across venues and time zones.
Can Agentic AI Reduce False Positives In Trade Monitoring?
Modern markets generate vast datasets across asset classes and jurisdictions, straining static monitoring frameworks. Rule-based systems can flood compliance desks with false positives, while more nuanced forms of manipulation may evade preset parameters. Agentic AI aims to examine multiple signals simultaneously, compare them with historical behavior, and escalate complex anomalies that merit human scrutiny.
Bloomberg reported that the systems are not intended to replace compliance officers but to function as an added analytical layer. The agents can assess relationships between trades, timing, market conditions, and trader history rather than isolated events, potentially improving detection accuracy. Yet accountability remains with human supervisors, particularly under regulatory regimes in the US and Europe that require explainable controls and audit trails.
The adoption reflects broader regulatory pressure to maintain effective market abuse monitoring systems. For banks, the operational appeal is clear: compliance teams face growing data volumes as trading becomes faster and more fragmented. If agentic AI can narrow alert volumes while preserving oversight standards, deployment is likely to expand as regulators review model governance and transparency frameworks in the next supervisory cycle.
