Web Stories Monday, November 16

Banks in the United Kingdom have drawn a line over how much access artificial intelligence should have to their customers’ money.

Four banks, including HSBC, Barclays, Lloyds, NatWest and Nationwide, have all moved to block or restrict third-party AI agents from accessing customer accounts, especially where they can make transactions without customer involvement.

There’s no evidence yet of AI agents accessing UK bank accounts just yet, but large lenders are worried about what happens if an automated system mismanages customers’ money — and who would be on the hook.

The banks might have other, unspoken concerns too.

Edgars Nemše, chief executive of GenLayer Foundation, an AI-focused blockchain infrastructure project, says routine banking tasks like paying bills, moving money and finding better rates could all be handed to agents.

Unfortunately for banks, this means agents would be able to seamlessly find customers a better deal elsewhere as soon as it’s published on the web.

“For plain accounts, that removes the switching costs and makes the space even more competitive,” Nemše tells Magazine.

The risk of agentic bank runs

That scenario could threaten more than just customer loyalty. In September, Apollo chief economist Torsten Slok warned about a possible “agentic bank run.”

If swarms of agents start moving household cash from low-yield checking accounts to alternatives offering higher returns en masse, that would drain banks of the low-cost funding they rely on to make loans. And the speed and scale “would be a problem for the entire financial system.”

Related: Coinbase CEO touts agentic finance as Base tops 100M AI payments

JPMorgan Chase chief executive Jamie Dimon has also warned against allowing external agents unrestricted access to bank accounts. “You don’t want rogue agents,” he said on an October earnings call.

Is an agentic bank run coming? Source: Apollo

Regardless of short-term moves to hold back the tide, AI agents look more and more likely to revolutionize banking within a few years.

Neeraj Makin, group head of strategy analytics at Emirates NBD, predicts agents will usher in a new era of around-the-clock banking and near-instant payments as normal by 2036.

“Forget about waiting for a payment for 24 hours or inter-country payments taking three days… all of that will be in seconds, atomic,” he says, with AI agents taking over most financial decisions.

Banks risk losing the everyday customer interface

Ferdinand Dabitz, chief executive of Augustus, an AI-native banking and payments company, sees a similar shift in how customers interact with financial institutions.

“Every consumer will have 10s and every business will have 100s of agents talking to their bank,” Dabitz says.

A human interface will still remain important for managing those agents and providing a layer of trust, “especially in commercial banking,” he says.

Related: Agents can trade 24/7 — but Yat Siu warns you could lose access at any time

Emirates NBD’s developer experience team argued in July that financial institutions should build their own frameworks for agentic AI, rather than allowing individual teams to develop agents without common controls.

Its Leap framework is designed to govern how agents authenticate with core banking systems, access sensitive data and carry out actions, with audit trails, monitoring and human oversight built in.

Emirates NBD says banks should build their own agentic frameworks. Source: Emirates NBD

However there are still major concerns around the reliability of agents to make major financial decisions on their own.

A recent Bank of England and FCA survey found that while 80% of responding firms had deployed AI agents, just 2% used agents with broad autonomy, identifying data security, third-party dependencies and risks like hallucinations among their biggest concerns.

From paying bills to making financial decisions

The AI takeover likely won’t happen overnight; “it will come in steps,” Nemše says.

“Today I use an agent to review my own and my company’s finances every month and queue the payments, and I approve them. The next step is the natural one: once you trust it, you let it spend small amounts without asking.”

Routine bills, subscriptions and the agent’s own running costs would be among the first tasks to become autonomous, he says.

Dabitz says corporate treasury management could be an area delegated to agents, but says banks will need to be involved in developing these early applications.

Who pays when an agent gets it wrong?

While consumers may enjoy the upside of automating their finances, who’s on the hook when an agent makes a mistake?

Nemše says banks need to offer “delegated access as a normal feature” to avoid customers sharing logins with agents. Customers could set permissions and spending limits, and agents could make low-value transactions independently while requiring approval for larger amounts.

But existing dispute processes may struggle if agents begin making thousands of transactions on behalf of each customer. “Even a small share going wrong is a lot of disputes,” he says. “The processes for handling them were built for people making a few payments a week. That part still has to be built.”

The US Office of the Comptroller of the Currency issued a supervisory framework for banks’ use of what it calls “super intelligence,” including agentic systems, on Nov. 4.

OCC’s Super Intelligence: Supervisory Framework. Source: OCC

The framework is meant to help bank examiners assess how institutions use the technology, and does not introduce new binding requirements for banks. Scrutiny should reflect the potential consequences of an agent’s actions, its level of autonomy and the degree of human oversight.

So an AI system helping a bank employee draft a report would present a different risk from one authorized to move customer funds without approval, and the banks that get that balance right are likely to come out on top.

Magazine: Bitcoin’s ‘best buying opportunity ever’ has arrived: Mark Moss

Read the full article here

Share.

Leave A Reply