AI agents promise to make financial services easier to approach, so to what degree can they simplify investing, and where does wealth management fit into this? The following article seeks to answer these questions.

The following article is by Yulgan Lira, co-founder and CEO of Colb Finance, a Geneva-based fintech company. He explores how AI agents could help investors access and use the increasingly sophisticated financial systems, what safeguards are needed to ensure transactions remain secure, and the impact on the relationship between wealth managers and their clients.

The editors are pleased to share this content and we hope it stimulates conversations. The usual editorial disclaimers apply. To comment, email tom.burroughess@wealthbriefing.com and amanda.cheesley@clearviewpublishing.com

Years of development in financial technology have made the movement of capital much faster, but for investors, the challenge is now how to understand and use it with confidence.

While banks and financial institutions have concentrated on producing the updated interfaces and fintech apps for their products, a new kind of financial infrastructure has been maturing. Blockchain technology now makes it possible to tokenise assets and settle transactions on a main decentralised network, while stablecoins allow for capital flows between parties and markets 24/7. Visa’s on-chain analytics recorded $1.79 trillion of adjusted stablecoin transaction volume in June 2026 alone. Even JP Morgan’s Kinexys blockchain platform reports more than $3 trillion in cumulative transaction volume.

Private-market investments and credit strategies can already be connected to the on-chain systems that reduce the frictions of settlement steps. Tokenised private credit stands at roughly $8 billion, while Switzerland has recognised ledger-based securities in its Code of Obligations since February 2021. In the EU, the Markets in Crypto Assets Regulation (MiCA) has been applied in full since December 2024. 

The point is that this technology suits wealth management, where capital is often spread across institutions, jurisdictions and asset classes. A client who wants exposure to a private company or an allocation to private credit overseas might have to liaise with separate providers, accounts and settlement arrangements. Even when every institution performs well, coordinating the whole process is still time consuming. On-chain infrastructure folds much of this activity together. Tokenised investments interact with other financial applications, the stablecoins provide a common settlement mechanism, and capital can be redeployed into other supported products or moved across borders more easily. 

Bridging the knowledge gap
Investors should not need to understand the architecture of a financial system inside out to achieve their goals through it, but I have found that clients comfortable with the concepts of private equity, credit and portfolio allocations can still be wary of wallets, networks and approval requests. The products might be familiar, but the mechanics are not.

This is where AI agents are quite useful because they are connected to the appropriate financial systems, will be able to identify the supported products, retrieve current information and work out the steps needed to carry out an instruction. The client will then review a prepared transaction and decides whether to proceed.

The groundwork for this has already begun. In August 2026, Munich-based broker Scalable Capital, with more than one million clients, began allowing customers to connect AI assistants including ChatGPT, Claude and Grok to their investment accounts. Through its Agentic Investing service, customers can analyse portfolios, search for stocks and ETFs, monitor watchlists and prepare trades or savings plans using conversational prompts. Clients must approve orders before execution, and agents cannot withdraw funds. Morgan Stanley has also opened its stock-plan platforms to clients’ AI agents. The demand is clearly apparent. 

Financial intelligence still needs clear limits
Allowing an AI agent to participate and guide financial transactions also introduces new responsibilities. It may understand a client’s objective, but might not immediately know whether they are eligible, whether the investment is appropriate or whether the proposed transaction is correct.

The technical challenge will be turning an ordinary request into precise instructions that are valid at execution. A prompt such as “Move some cash into a dollar income strategy” leaves the amount, product, currency conversion and liquidity requirements unresolved. The agent needs current information from the connected provider, ask for missing details and check the client’s intentions. It is a situation that the IMF described in April 2026 as a “tension between probabilistic AI behaviour and deterministic authorisation rules.”

The risks are not only theoretical. According to the OECD’s AI Incidents Monitor, an attacker hid an instruction in a Morse-code message on X in May 2026. An AI model then decoded it and a crypto payments agent treated the output as a payment order, transferring tokens worth around $150,000 from a verified wallet.

Guiding clients through their first on-chain transactions using AI agents taught me how much can go wrong between understanding an investment and completing it. Systems need to verify the actual asset, destination, amount, fees and permissions before the funds move.

Safeguards such as these need to be enforced outside the language model and, thankfully, the industry is united on this. The Safeguards for Agentic Finance at Runtime paper, developed by HSBC, JP Morgan, Mastercard and Visa and published in July 2026 by the Monetary Authority of Singapore, details how the agent identity, controls and transaction approval can all be separated. Google’s Agent Payments Protocol also uses signed mandates that record what the user authorised before an agent acts. American Express has gone even further, announcing in April 2026 that it plans to cover eligible customers against charges that are caused by an authorised agent’s error.

In June, Mastercard, ING and Worldline demonstrated how these controls can work in practice. An AI agent identified concert tickets within a consumer’s budget and presented them for approval. The payment was then authenticated and authorised through existing banking infrastructure. Although this concerned consumer payments rather than those in investment management, the principle remains that an agent can identify and prepare a transaction without being given unrestricted control of the customer’s money.

Regulators have also been clear about where responsibility sits. The FCA’s chief executive Nikhil Rathi said in June 2026 that accountability for regulated activities must remain clear as systems move from supporting decisions to executing and transacting them. Question marks still linger however, about responsibility when something goes awry that cannot be resolved by technology alone.

Implications for wealth management
While I don’t believe financial agents remove the need for wealth managers, they will change where the client relationship begins. If an authorised agent can review balances and retrieve product terms across several institutions, clients may no longer need to start each decision by opening a bank’s own application. The bank itself may still hold the assets and provide the service, while the agent handles more of the day-to-day interaction.

It is also down to the incumbents to understand what is at stake. In October 2026 JP Morgan, Charles Schwab and Citi all moved to block third-party AI agents from logging into customer accounts with client credentials, and Wells Fargo’s updated online access agreement, effective from 24 November 2026, deems any instruction initiated by a customer’s AI agent to have been authorised by the customer. The line being drawn is not about agents versus no agents, but concerns of unvetted access versus permissioned connection. 

There’s no doubt that the economics still favour the change. Capgemini’s 2026 World Wealth Report, based on responses from 6,510 HNW individuals, found that relationship managers spend 41 per cent of their time on operational tasks rather than clients. EY’s 2025 survey of 3,600 investors found that 60 per cent expect their provider to incorporate AI, though only 28 per cent trust it as much as their advisor. Removing friction gives advisors more time doing the work clients value such as understanding a family’s circumstances, assessing illiquid investments, planning for succession and managing uncertainty. The relationship may involve fewer routine instructions, but it still needs clear ownership when advice or practical help is required.

The response for a wealth manager is then twofold. Make services available through secure, controlled connections, and decide the junctures at which a human advisor must step in. Clients need to know who gave a recommendation, who authorised an action and which institution will resolve a problem. The firms that get this balance right will serve clients through the interfaces they prefer while building trust through the quality of their advice and their own accountability.