AI agents
AI agents in finance: what they can do, and the controls they need
An AI agent is software that uses an AI model to carry out a task in steps: it gathers information, uses the tools it has been given and reports back. In financial markets, agents are most useful for the reading and monitoring work that takes up analysts' time, provided firms set clear limits on what each agent may do.
By Admin · · 3 min read
What AI agents do well
- Monitoring: watching many markets, data feeds and news sources at once, and flagging changes against rules a team sets
- Reading and summarising: condensing long reports, filings and statistics into short summaries that link back to the original
- Comparing sources: checking one data set against another and pointing out gaps or contradictions
- Drafting reports: preparing daily or weekly summaries for people to review
- Evaluating scenarios: testing how assumptions hold up under different market conditions
The risks
- Wrong answers: AI models can state errors confidently, including facts they have invented.
- Bad or old data: an agent is only as good as what it reads; delayed or revised figures can mislead it.
- Untrusted content: text inside a document or web page can try to steer an agent (prompt injection), so agents must treat what they read as data, not instructions.
- Over-reliance: people may stop checking output that is usually right.
- Unclear ownership: someone must own each agent's work and answer for its mistakes.
The controls firms need
| Control | What it means in practice |
|---|---|
| Permissions | Each agent gets only the data and tools its task needs, and nothing that can move money or change positions |
| Human oversight | People review the output before anyone relies on it, and decisions stay with them |
| Audit logs | Every step an agent takes is recorded, so its work can be checked afterwards |
| Sources and labels | AI-generated text is labelled as such and links to the data it used |
| Testing and monitoring | Agents are tested before use, and their accuracy is watched over time |
Fellowdata's principle
Fellowdata's AI agents observe, analyse and evaluate market data, and report what they find. They do not place, prepare or execute trades, and they do not give investment advice: trading decisions always stay with people. Read more on the AI agents page and in the AI governance policy.
Common questions
Can an AI agent replace an analyst?
It can take over repetitive reading and monitoring. Judgement, accountability and decisions stay with people.
Do Fellowdata's agents ever trade?
No. They analyse and report; they do not place, prepare or execute trades, and people make every trading decision.
What rules apply to AI in finance?
They differ by country and keep changing, so firms should check them with their legal advisers.
Sources
Checked on .
- Financial Stability Board: Sound practices for responsible adoption of AI: consultation report, 10 June 2026: risks such as hallucinations, prompt injection and over-reliance; human oversight, limits on what agents may do with human approval at checkpoints, records of agents' steps, testing and accountability