Goldman Sachs and Deutsche Bank Test Agentic AI for Trade Surveillance

Global investment banks are entering a new phase of artificial intelligence adoption — one that moves beyond chatbots and predictive dashboards into the core of market oversight.

Industry leaders like Goldman Sachs and Deutsche Bank are now testing agentic AI systems to enhance trade surveillance. Unlike traditional rule-based monitoring tools, these advanced AI agents are designed to reason through trading patterns in real time, identify subtle misconduct risks, and escalate concerns for human review.

This shift could redefine how compliance teams monitor market abuse, insider trading, spoofing, and other regulatory risks in increasingly complex global markets.

But what exactly is agentic AI in trading surveillance? Why are major banks investing in it? And what does this mean for compliance teams, regulators, and the future of financial oversight?

Let’s break it down.


The Limits of Traditional Trade Surveillance Systems

For years, banks have relied on automated trade surveillance systems built around predefined rules.

These systems typically function like this:

  • If a trade exceeds a size threshold → trigger alert
  • If pricing deviates from benchmark → trigger alert
  • If activity matches a known manipulation pattern → trigger alert

Compliance officers then manually investigate each alert to determine whether misconduct occurred.

While effective in principle, rule-based surveillance faces growing challenges in modern markets.

1. Exploding Data Volumes

Global markets generate massive amounts of trading data across:

  • Multiple asset classes
  • Numerous trading venues
  • Various time zones
  • High-frequency trading environments

The scale alone makes static rule systems increasingly inefficient.

2. High False-Positive Rates

Rule-based systems often generate overwhelming numbers of alerts. Many are false positives — technically suspicious by rule definition but harmless in practice.

Compliance teams spend significant time reviewing low-risk cases, reducing efficiency and increasing operational costs.

3. Evolving Manipulation Tactics

Market misconduct strategies evolve. Traders attempting manipulation rarely follow predictable templates. Subtle combinations of timing, communication signals, and order placement may evade simple rule detection.

Static systems struggle to catch patterns that do not precisely match known red flags.


Enter Agentic AI: A Smarter Surveillance Model

Agentic AI introduces a more dynamic approach to monitoring trading activity.

Instead of merely checking trades against fixed criteria, AI agents can:

  • Analyse multiple data signals simultaneously
  • Compare behaviour against historical activity
  • Identify unusual combinations of actions
  • Adapt investigative focus based on emerging patterns

The term “agentic” refers to systems capable of goal-directed reasoning rather than passive responses.

In a trading context, that means AI agents can:

  • Decide which data to evaluate next
  • Examine contextual relationships
  • Connect patterns across accounts and timeframes
  • Escalate findings when risk thresholds are exceeded

This transforms surveillance from reactive alerting into proactive behavioural analysis.


Deutsche Bank’s Collaboration with Google Cloud

According to reporting, Deutsche Bank is working with Google Cloud to develop AI agents for monitoring trading activity.

The system aims to review extensive sets of:

  • Order data
  • Execution records
  • Trade timestamps
  • Market condition metrics

By analysing structured and unstructured data in near real time, the AI agents can flag anomalies that may not be evident through traditional systems.

Importantly, the system is designed to detect “complex anomalies.” This suggests it evaluates relationships between:

  • Trader history
  • Order flow patterns
  • Market volatility
  • Cross-asset activity
  • Timing clusters

Rather than examining a single trade in isolation, the AI considers behavioural context.

However, the system does not replace compliance teams. Human supervisors remain responsible for reviewing flagged cases and determining appropriate action.


Goldman Sachs’ Agentic AI Strategy in Surveillance

Goldman Sachs has also been exploring agentic AI tools for enhancing its surveillance infrastructure.

The bank has invested heavily in AI across trading, risk modelling, and operational systems in recent years. Expanding AI capabilities into compliance is a natural progression.

The agentic approach reportedly focuses on identifying patterns that may not violate explicit rules but still appear anomalous.

For example:

  • Unusual coordination between traders
  • Suspicious timing around earnings announcements
  • Patterned order placements near price-sensitive events
  • Deviations from a trader’s historical strategy profile

By using AI agents capable of contextual reasoning, the bank aims to strengthen early detection capabilities.


Why Real-Time Pattern Recognition Matters

Financial markets operate at extraordinary speed. In high-frequency environments, suspicious behaviour can unfold in milliseconds.

Rule-based systems typically detect misconduct after thresholds are crossed. Agentic AI systems, by contrast, can evaluate behavioural signals in motion.

This shift enables:

  • Earlier detection of market abuse
  • Faster escalation to compliance teams
  • Reduced reputational damage
  • Lower regulatory risk exposure

For regulators, improved detection supports market integrity. For banks, it reduces potential fines and enforcement actions.


The Regulatory Environment Driving Innovation

Financial institutions operate under strict regulatory frameworks in the United States, Europe, and globally.

Regulators require firms to maintain “effective systems and controls” to detect and prevent market abuse.

While regulators do not mandate agentic AI specifically, they expect surveillance systems to evolve alongside market complexity.

As trading volumes grow and misconduct tactics become more sophisticated, institutions must demonstrate that their oversight mechanisms remain robust.

AI-powered surveillance may help firms meet that standard.

However, adoption also introduces new compliance challenges.


Governance and Explainability Concerns

Deploying advanced AI systems in compliance functions raises critical questions:

  • Can the AI’s reasoning be explained?
  • Is there an audit trail for decisions?
  • Are outputs reproducible?
  • Does the model introduce bias?
  • How is data secured?

Financial regulators demand transparency.

If an AI agent flags suspicious activity, compliance officers must be able to understand:

  • Why it was flagged
  • Which signals contributed
  • How confidence levels were calculated
  • Whether the reasoning aligns with regulatory definitions

Without explainability, AI risks becoming a liability rather than an asset.

Therefore, governance frameworks remain central to successful implementation.


Agentic AI Does Not Replace Human Judgment

Despite its advanced reasoning capabilities, agentic AI does not make disciplinary decisions.

Human compliance professionals:

  • Review AI-generated alerts
  • Conduct deeper investigations
  • Engage with traders
  • Escalate to legal teams
  • Interface with regulators

AI acts as an intelligent filter and pattern recogniser — not a decision-maker.

This human-in-the-loop structure ensures accountability remains with licensed professionals.

In highly regulated industries, such oversight is non-negotiable.


A Broader Shift in Compliance Technology

The use of agentic AI in trade surveillance reflects a wider transformation in financial compliance.

Historically, compliance systems focused on:

  • Static rules
  • Manual audits
  • Periodic reporting

Modern compliance increasingly demands:

  • Continuous monitoring
  • Data-driven anomaly detection
  • Cross-platform behavioural analysis
  • Real-time escalation

Generative AI architectures and large language models are now being adapted for internal control functions rather than customer-facing applications.

This marks a significant evolution in enterprise AI usage.


Reducing Alert Fatigue in Compliance Teams

One major operational benefit of agentic surveillance systems is the potential reduction in alert fatigue.

Traditional rule-based systems can overwhelm teams with thousands of low-risk alerts.

Agentic AI aims to:

  • Reduce false positives
  • Prioritise higher-risk cases
  • Consolidate related signals
  • Present contextual summaries

This allows compliance staff to focus on complex investigations rather than routine screening.

In an environment where skilled compliance professionals are in high demand, efficiency gains are strategically valuable.


How Agentic AI May Reshape Compliance Roles

If agentic surveillance tools prove effective, compliance roles may evolve.

Rather than spending most of their time triaging simple alerts, teams may focus on:

  • Evaluating nuanced behavioural patterns
  • Interpreting AI reasoning traces
  • Engaging in preventative risk management
  • Strengthening governance controls

Human expertise shifts from detection to interpretation and strategic oversight.

This does not eliminate jobs. Instead, it elevates the skill requirements within compliance departments.


Balancing Innovation with Responsibility

Adopting agentic AI for trade surveillance offers compelling advantages:

  • Enhanced detection accuracy
  • Faster response times
  • Improved regulatory compliance
  • Reduced operational costs

But it also demands careful management.

Banks must ensure:

  • Strong model governance
  • Robust cybersecurity protections
  • Transparent audit capabilities
  • Ongoing performance evaluation
  • Clear accountability structures

The integration of AI into compliance cannot compromise regulatory standards.


The Competitive Advantage of Intelligent Surveillance

For major institutions like Goldman Sachs and Deutsche Bank, adopting agentic AI is not only about regulatory compliance.

It is also about competitive positioning.

Firms that detect misconduct early can:

  • Avoid costly enforcement actions
  • Protect brand reputation
  • Strengthen client confidence
  • Demonstrate proactive governance

In global finance, trust is currency.

Agentic AI may become a strategic differentiator in maintaining that trust.


The Future of AI in Financial Market Oversight

The deployment of agentic AI in trade surveillance signals a broader industry trend:

AI is moving from support function to strategic infrastructure.

As markets grow more data-intensive and complex, the ability to reason across signals in real time becomes essential.

Static rule engines alone may no longer suffice.

Agentic AI systems capable of adaptive analysis, contextual reasoning, and real-time pattern detection represent the next frontier in compliance technology.

However, success will depend not only on technological sophistication but also on governance, transparency, and human oversight.


Final Thoughts

The experimentation by Goldman Sachs and Deutsche Bank with agentic AI for trade surveillance marks a pivotal moment in financial regulation technology.

Rather than replacing compliance professionals, these AI agents aim to strengthen oversight by reducing noise, identifying subtle anomalies, and accelerating detection timelines.

As regulatory scrutiny intensifies and market complexity increases, intelligent surveillance systems may become indispensable.

But the ultimate measure of success will not be how autonomous these systems become.

It will be how effectively they combine advanced reasoning with accountable human judgment — ensuring that innovation enhances, rather than undermines, market integrity.