Goldman, Deutsche Bank Pilot Agentic AI for Surveillance
Goldman Sachs and Deutsche Bank are testing next-gen agentic AI systems for real-time trade surveillance. These agents go beyond static rules to dynamically reason through trading patterns, flagging compliance risks that legacy systems would miss. It's a major sign that banks are now using AI for context-aware decisions, not just workflow automation.
Deutsche Bank is working with Google Cloud to develop its agentic AI surveillance system. This initiative includes plans to deploy a large language model in 2026 to monitor communications from traders and other client-facing staff, looking for anomalies in orders and trades. As part of a broader compliance overhaul, the bank has already retired approximately 200 legacy servers previously used for surveillance. Goldman Sachs has been co-developing autonomous AI agents with the AI company Anthropic, utilizing its Claude model. Their CIO, Marco Argenti, described the agents as a "digital co-worker" for complex, process-intensive roles, with employee surveillance being a specific target for this technology. This move represents a shift from automating trading execution to automating internal compliance functions. These AI agents move beyond static, predefined rules that generate a high volume of false positives. Instead, they can autonomously decide which data to analyze, compare multiple signals like order flows and communication metadata against historical behavior, and escalate unusual patterns without constant human input. This allows for the detection of more subtle or complex forms of potential misconduct that legacy systems might miss. The adoption of this technology is part of a broader industry trend. Nomura is reportedly in talks with another global bank to jointly train AI surveillance models and is even discussing potential funding with regulators. Meanwhile, Banco Santander is working with fintech firm ThetaRay to implement agentic AI for its anti-money laundering (AML) controls. This shift is driven by the sheer scale and complexity of modern financial markets, which generate massive volumes of data across numerous asset classes and venues. The global market for trade surveillance technology was valued at approximately $1.7 billion in 2024 and is projected to grow to $5.2 billion by 2030, highlighting the significant investment firms are making to enhance their compliance capabilities. Agentic AI systems in finance combine multiple autonomous agents that can orchestrate complex workflows. For instance, one agent might ingest order data and chat logs, while a planner agent directs specialized agents to test hypotheses against historical patterns, with human analysts retaining final authority over decisions. Some analyses suggest this approach could automate up to 70% of manual compliance work and improve the accuracy of risk detection by a factor of four.
