Goldman Sachs and Deutsche Bank test agentic AI to strengthen trade surveillance

Global investment banks are beginning to test a new generation of “agentic” artificial intelligence for trade surveillance, moving beyond static, rule-based systems. According to a Bloomberg report, Goldman Sachs and Deutsche Bank are exploring AI agents that can analyse trading behaviour in real time and flag complex patterns of potential misconduct for human review.

Major global banks are experimenting with agentic artificial intelligence to enhance oversight of trading activity, signalling a shift in how compliance and surveillance functions may operate in the future. According to Bloomberg, Goldman Sachs and Deutsche Bank are testing or deploying AI systems designed to reason through trading patterns dynamically, rather than relying solely on preset rules and keyword-based alerts.

Traditionally, automated trade surveillance systems operate on static thresholds — flagging transactions that exceed certain sizes, deviate from benchmarks, or match known risk scenarios. While effective at scale, these systems often generate large volumes of false positives and can miss more subtle forms of market manipulation that do not conform to predefined patterns.

Agentic AI systems aim to address those limitations. Instead of simply matching trades against a checklist, these AI agents can analyse multiple signals simultaneously, compare activity with historical behaviour, and identify unusual combinations of actions as trades occur. The tools are intended to support, not replace, human compliance officers by surfacing cases that warrant closer inspection.

Deutsche Bank’s Work With Google Cloud

Bloomberg reported that Deutsche Bank is collaborating with Google Cloud to develop AI agents capable of monitoring orders and executions across large datasets in near real time. The system is designed to identify “complex anomalies” by examining relationships between trades, timing, market conditions, and trader history, rather than assessing transactions in isolation.

The initiative reflects Deutsche Bank’s broader push to apply generative AI and large language model technologies beyond customer-facing applications. In this case, AI is being used to analyse structured and unstructured trading data streams to strengthen internal controls. Human compliance teams remain responsible for reviewing flagged activity and determining next steps.

Goldman Sachs’ Agentic AI Strategy

Goldman Sachs is also exploring agentic AI as part of its wider investment in AI-driven trading, risk, and compliance systems. According to the report, the bank is testing AI agents that operate with a degree of autonomy in scanning for indicators of misconduct, including patterns that may not clearly violate specific rules but appear abnormal when viewed holistically.

For banks, the appeal lies not only in stronger oversight but also in efficiency. Compliance departments face growing pressure to manage increasing data volumes while maintaining rigorous standards. AI agents that can reduce alert noise without lowering scrutiny could significantly change how compliance teams allocate their time.

Why Agentic AI Matters

Agentic AI refers to systems capable of goal-directed actions, such as deciding which data to analyse next, correlating multiple signals, and escalating findings without constant human prompts. In a trading environment, this can involve monitoring order flows, price movements, and behavioural history to assess whether activity aligns with expected patterns.

Importantly, these systems do not make disciplinary decisions. Financial institutions remain accountable under strict regulatory regimes, and human oversight is mandatory. The role of agentic AI is to improve detection and organisation of information, not to replace judgement.

A Broader Shift in Compliance

The use of agentic AI in trade surveillance reflects a wider trend of applying advanced AI architectures to internal control functions. Regulators in the US and Europe have encouraged firms to strengthen market abuse monitoring, and while agentic AI is not mandated, effective systems and controls are required.

At the same time, AI-driven compliance introduces new challenges around explainability, bias, data security, and auditability. Banks must ensure that AI models can withstand regulatory scrutiny and that governance frameworks keep pace with technological capability.

If these tools prove effective, they could reshape compliance workflows — shifting human effort away from processing large volumes of simple alerts toward evaluating more complex cases surfaced by AI. In increasingly fast and data-intensive markets, such capabilities may become a critical component of effective trade surveillance.