AI agents are a risk to financial stability

More than a chatbot

Don't simply call an AI agent a chatbot. A chatbot answers your question. An AI agent is a digital assistant that not only provides answers but also carries out tasks independently, within the instructions and permissions you give it. Big Tech is going all in. Meta launched Muse in the US and Canada on September 8, OpenAI announced 'dots' on September 29, Microsoft is coming out with Autopilot and Google with Gemini Agent.

A revolution in the workplace

AI agents are as revolutionary as apps were in their day. They are making many jobs at large companies obsolete, including highly skilled ones. Banking giant HSBC, for example, announced last week that it plans to cut around half of the management and specialist roles in its UK wealth management division, as well as 70% (!) of its financial advisors. Small business owners and freelancers are also suddenly seeing a sea of opportunities open up. Finally, a secretary! What's more, you can build your own huge team of AI agents working for you at a fraction of the cost of human employees. AI agents never take vacations and – although you can't rule it out entirely – have not yet set up a virtual labor union. Yes, they (still) make mistakes, but that is improving very quickly within AI's self-learning systems.

The end of financial inertia

AI agents can also shift the balance of power between banks and their customers. Many people are financially inert. They complain about low interest rates on their savings but are often too complacent to do anything about it themselves. That inertia is a blessing for banks, which can build a model of cheap deposit funding on it. This is certainly the case in Belgium, although those low savings rates should be seen in relation to the very favorable mortgage rates. An AI agent can break through that inertia. Savers can use AI agents to put their savings to work by instructing them to continuously track down the bank offering the best interest terms, taking into account taxes, deposit protection, conditions and any loss of loyalty bonuses. Give the AI agent the authority to move your savings elsewhere on its own, and you're all set.

Lessons from Silicon Valley Bank

AI agents are also redrawing the risks in the banking landscape. The far-reaching digitalization of the financial economy has many advantages, but it also increases the risk of bank runs. The collapse of Silicon Valley Bank in 2023 illustrates this. The bank got into trouble because it suffered heavy losses on its portfolio of US Treasury bonds, which had fallen sharply in value as interest rates rose. As long as customers did not withdraw their deposits, these losses were only latent and the bank could try to ride it out. But once customers began withdrawing deposits en masse, the bank had to sell the bonds to raise cash, and the losses were realized. After a few unfortunate statements by the CEO about the bank's financial position, a tech mogul lit the fuse by urging others on social media to pull their money out. And unlike in the pre-digital era, that can be done in the blink of an eye via an app. The digital bank run brought Silicon Valley Bank down in less than two days.

Speed and simultaneity: the bank run of the future

At Silicon Valley Bank, customers still had to press the button themselves, but with an AI agent even that is no longer necessary. You can also instruct your AI agent to pull your money out at lightning speed as soon as bad news emerges. The danger lies in both speed and simultaneity. The next bank run may not wait for savers to wake up. Millions of agents can process the same information and follow similar decision rules. They don't need to coordinate anything to head for the exit at the same time. Even a rumor can trigger a self-reinforcing process: withdrawals lead to liquidity problems, which in turn prompt further withdrawals.

Pressure on interest margins and lending

At the end of September, Torsten Sløk, chief economist at Apollo, also pointed out that many fintech companies offer considerably higher interest rates than US banks. Revolut Standard, for example, offers 3.5% interest, a full percentage point more than US banks offer on a regulated savings account. Stablecoin issuers (stablecoins are cryptocurrencies pegged to the dollar) are not yet allowed to offer interest, but sooner or later that will happen. If banks raise their deposit rates across the board, they will have to raise their lending rates if they want to protect their interest margin (the difference between the interest they pay on deposits and the interest they charge on loans). That too is a systemic risk, as it puts pressure on economic growth.

How banks and regulators can respond

Banks can, of course, push back against this risk. They can, for example, raise their interest rates, cap the incoming amounts that can earn preferential rates and put up digital barriers to slow AI agents down. But that doesn't change the fact that AI agents increase systemic risk. Banks and regulators will need to adapt their stress tests to that possibility. And new regulation may well follow.

Physical gold outside the banking system

AI agents show how quickly money can move within the digital financial system and how existing risks can be amplified as a result. In that respect, a bank balance and physical gold have different risk profiles. Money in an account remains dependent on the functioning of the banking system, whereas physically allocated gold is held outside it. Precisely now that financial processes are becoming ever more digital and ever faster, gold can therefore play a complementary, diversifying role within a well-spread portfolio.