AI agents in capital markets
AI agents in capital markets are software systems that plan and carry out several steps in securities-related work. Agents use tools such as market data feeds, order systems and case management software to gather information and act on a task. Trading, surveillance, research and post-trade processing require different permissions and levels of human oversight.
Capital market teams use AI agents to connect tasks that would otherwise require someone to move information between systems. An agent can assemble the material for a review and hand the result to the person responsible for a decision. Firms need traceable actions and clear approval limits so that automation remains compatible with their responsibilities to clients and the market.
Agents in trade surveillance
Deutsche Bank is working with Google Cloud on a large language model that flags anomalies in orders, trades and market moves, Bernd Leukert, its head of technology, data and innovation, told Bloomberg. Goldman Sachs is also exploring agentic AI to detect possible trader misconduct. A surveillance agent can collect the orders, messages and market data around an alert and draft the case for the compliance officer.
The decision on a suspicious transaction report stays with a person. The trade surveillance in Germany page covers the duties under the Market Abuse Regulation.
Research and client onboarding
Data providers have opened their systems to agents. LSEG runs an MCP server for its licensed data, so an analyst's agent can pull prices and company data into its work. Goldman Sachs built agents on Claude from Anthropic for client vetting and onboarding, according to InvestmentNews.
The research firm Celent calls agentic AI adoption in capital markets early. Celent expects agent chaining, in which specialized agents each handle one part of a workflow, to raise straight-through processing, and advises firms to plan for agentic workflows within about five years.
The rules in the EU
ESMA's survey of AI in EU securities markets found that agentic use cases were about a sixth of all reported cases and ran with high or medium autonomy more often than other AI. An agent that sends orders falls under the MiFID II rules for algorithmic trading, which require testing, a kill function and records. An agent that monitors trading supports the firm's duty under the Market Abuse Regulation to detect and report suspicious orders.
The AI in financial trading page covers trading models, and the EU AI Act in finance page covers the AI rules for financial firms.
Upcoming AI and finance events in Germany
What are AI agents in capital markets?
They are AI systems that plan several steps and use market data, order and case management systems as tools, for tasks such as trade surveillance, research and client onboarding.
How do AI agents differ from trading algorithms?
A trading algorithm follows fixed rules: when a market condition occurs, it runs a coded response. Chainlink describes AI agents as software that reasons and decides on the next step with little human input.
Which banks use AI agents in capital markets?
Deutsche Bank and Goldman Sachs work on agentic AI for trade surveillance, and Goldman Sachs uses agents for client vetting and onboarding.
AI agents in capital markets and Finance Loop
Finance Loop brings surveillance and trading teams in Frankfurt together with the developers who build agents for them. Finance Loop covers the topic in its Investment & Digital Assets and Risk & Compliance tracks. Market data and surveillance vendors can present in a sponsored webinar or a meetup, and practitioners join through the membership.
Finance Loop is a professional network supporting the institutional use of emerging technologies in finance, such as AI, digital payments, cloud and blockchain solutions. Finance Loop helps its members build skills and personal networks in these fields: Investment & Digital Assets, Payments & Digital Money, Digital Infrastructure & Sovereignty, and Risk & Compliance.