Would You Trust an Algorithm With Your Savings? Financial Agents Examined
An automated agent feels neither fear nor greed, yet it may err alongside millions of others at once. What do financial agents promise, and what can they not do?

The Nano capsuleOne minute
An AI financial agent does not merely advise; it may carry out decisions within limits its owner sets. What it most plainly targets is investor behaviour itself: Morningstar found that investors in US funds earned 8.7% a year over the ten years to the end of 2025, while the funds returned 9.9%, the difference being the timing of purchases and sales. Yet the Financial Stability Board warned in 2024 that similar tools could push markets in the same direction at the same time. In March 2026 the board of Saudi Arabia's Capital Market Authority approved draft rules for robo-advisory services. The agent is a tool that helps; it does not take over the owner's responsibility.
Read the full article Collapse the article
Who guards your savings while you sleep? Not your bank, which closes in the evening, but perhaps a programme watching the markets around the clock. That is the promise of the "financial agent": a system that grasps your goal, then plans and acts within the powers you grant it in advance.
From calculator to agent
The first wave was the "robo-advisor", a programme that allocates and rebalances a portfolio according to fixed rules. Agents built on large language models are meant to go beyond rigid rules: to read the owner's circumstances and act within the mandate given. That idea remains more promise than measured reality; there is as yet no published evidence that agents improve individuals' returns. For an introduction to the models, see the article "How Do Large Language Models (LLMs) Like ChatGPT Work?".
The problem the agent tries to solve
The clearest problem in wealth management is that investors mistime the market themselves. Morningstar measures the difference between what funds earn and what investors in them actually take home. Over the ten years to 31 December 2025, US funds returned 9.9% a year and the average investor 8.7%: a gap of 1.2 percentage points a year, about 12% of total return, which Morningstar attributes to the timing and size of purchases and sales. For more on this gap and on the robo-adviser, see the article "The Robo-Adviser: Can an Algorithm Guard You from Yourself?".
That suggests automated discipline might narrow the gap, since a programme does not panic in a fall or get carried away in a rise. But it is a plausible hypothesis, not a proven result; an investor who does not understand the agent may cancel the mandate at the worst moment.
Three risks regulators point to
In November 2024 the Financial Stability Board published its report on AI and financial stability, identifying among the vulnerabilities:
- Similarity: if many institutions adopt similar tools, markets may move in the same direction at the same time, amplifying disturbance instead of dissipating it.
- Dependence on providers: when a service is concentrated in a few hands, one outage can spread.
- Cyber risk, and model and data quality: a poorly governed model can produce wrong decisions with confidence.
There is also a question the report alone does not answer: to whom does the agent owe loyalty, the owner of the money or its maker? The practical answer is disclosure and accountability, which is the direction the rules take.
The Saudi rules and the Sharia consideration
According to Argaam, the board of the Capital Market Authority approved on 5 March 2026 draft rules for robo-advisory services, following trials by licensed firms in the FinTech regulatory sandbox. Among the main points reported by Argaam and the law firm Al Tamimi:
- The service may be offered only through firms licensed in investment management.
- The Authority must be notified of portfolio strategies in advance, and algorithms tested periodically; Al Tamimi adds a requirement for annual independent safety testing.
- The platform must disclose the role of the algorithm, how it works and its risks.
- Historical performance since inception must be published, net of actual expenses.
- The IT officer who runs the systems must be registered.
The two sources differ in how they describe the step, whether as amendments approved or a draft; consult the Authority's official text when it matters.
It matters to a Muslim that an investment conforms to Sharia, avoiding riba and guarding against gharar. An algorithm may screen instruments against set criteria, but the Sharia ruling comes from the competent Sharia body, not from the programme. So confirm that the product has been approved by a recognised Sharia board before you delegate.
Questions before you delegate
- What limits can I set, and can I stop the agent at once?
- Is the firm licensed, and what does it disclose about how decisions are made?
- Who bears the loss if the system errs?
- Do I understand what it does well enough to review it regularly?
Wealth is a trust, and managing it with knowledge and care is part of taking the means; delegating to a machine does not remove the owner's responsibility. This material is general knowledge, not an investment recommendation; consult a licensed professional.
References
- Morningstar, Mind the Gap: A Report on Investor Returns in the US (10 years ended 31 December 2025)
- Financial Stability Board, The Financial Stability Implications of Artificial Intelligence (press release, 14 November 2024)
- Argaam, CMA approves regulation of robo-advisory services (March 2026)
- Al Tamimi & Company, Capital Market Authority approves the regulation of robo-advisory services: key takeaways (13 March 2026)

