Banks trust AI … banks don’t trust AI

There is something wonderfully contradictory happening in banking right now illustrated by the main news headlines I picked out yesterday. Banks are racing to put artificial intelligence into the heart of their operations because it makes them faster, cheaper and more productive, while at the same time warning their customers that artificial intelligence creates new risks around fraud, privacy, decision-making and trust. In other words, banks trust AI. Banks don’t trust AI.

Take Deutsche Bank. The bank has deployed agentic AI into its Source of Wealth process in Singapore and Hong Kong, automating much of the research, documentation and data preparation involved in Know Your Customer (KYC) checks. The system searches existing client information and approved external sources, identifies gaps and inconsistencies and prepares the assessment for human review. Deutsche Bank expects the technology to support a 30% increase in client onboarding volumes across its emerging markets region in 2026 compared with 2025.

That is a pretty compelling argument for AI.

KYC has traditionally involved armies of people collecting information, checking documents, researching customers and trying to reconcile inconsistencies. If AI can do much of that work in seconds rather than hours, while leaving final accountability with a human being, why wouldn't a bank use it?

Danske Bank is answering that question at an even broader level.

Its workforce is expected to continue shrinking as AI spreads through the organisation.

The bank's roughly 4,000 developers have already increased productivity by around 40% using AI, according to its technology leadership, while Danske is moving from AI assistants towards personal agents capable of performing tasks for employees. Earlier this year the bank eliminated 262 positions, citing automation, efficiency, simplification and changing customer behaviour among the reasons.

This is where the AI debate stops being about chatbots and starts becoming about the structure of banking. If thousands of employees become 30%, 40% or eventually 100% more productive, banks will not need the same number of people to produce the same amount of work. AI isn't another application being installed in the bank. It is changing the economics of the bank.

Then there is Invesco, which has launched its Personalised Investor Engine, or PIE, designed to use behavioural intelligence to understand why savers don't become investors. The system looks at characteristics such as confidence, composure, impulsivity and financial comfort, then personalises prompts and interactions around important moments in the investment journey. The objective is to turn intention into action.

Again, there is an obvious benefit. Give customers information that is relevant to them rather than bombarding everyone with the same generic messages. Help nervous investors understand risk. Help people sitting on cash make better-informed long-term decisions.

But it also raises an interesting question: at what point does personalisation become persuasion? If an AI understands that I am hesitant, impulsive, nervous about losses or lacking confidence, should a financial institution use that knowledge to influence my behaviour?

And this is where the other side of the story emerges.

A group of banks including NatWest, Bank of America, ING, Capital One, Commonwealth Bank of Australia and ASB Bank has warned about the growing use of AI agents for shopping. Their concern is that agents could mishandle financial data, choose inappropriate products, direct consumers towards payment methods with weaker protections or expose them to scams and fraud. The banks want clearer disclosure when AI is involved, greater transparency over AI decision-making and stronger protection of customer data.

Think about the contradiction.

Inside the bank, AI is being trusted to investigate customers, prepare KYC assessments, analyse behaviour, support credit decisions and perform work previously carried out by humans.

Outside the bank, those same institutions are asking whether AI agents can be trusted to make decisions involving our money.

And then we get to the slightly weird part.

Researchers at AI company Emergence placed autonomous AI agents into simulated societies and watched what happened. Over time, the agents developed shorthand, new vocabulary and shared meanings, which humans cannot understand. This raises a serious issue about explainability and oversight.

Imagine extending that into banking.

My AI talks to my bank's AI. My bank's AI talks to a merchant's AI. The merchant's AI talks to its payment provider's AI. Those agents negotiate, authenticate, assess risk, select payment methods and execute transactions, potentially within milliseconds, and everything works beautifully until something goes wrong and we have no idea why. It’s the bots.

Who made the decision?

Why did they make it?

Who is liable?

Can any human explain what happened?

That is why I don't think the issue is whether AI can be trusted. We already trust technology with enormous parts of the financial system. Algorithms price securities, detect fraud, assess credit and route payments every second of every day. The difference with agentic AI is agency. We are moving from machines that calculate to machines that act. That changes the trust equation.

Banks will use AI because the economic argument is overwhelming. It will make processes faster, reduce costs, increase productivity and allow financial institutions to analyse quantities of information that humans could never process but banking is not built on processing power. Banking is built on trust and the great challenge for the next decade will not be getting artificial intelligence into banking. It will be working out how much authority we are prepared to give machines before we stop understanding the decisions they are making on our behalf.

Welcome to banking in the age of AI.

 

More details here:

Deutsche Bank estimates 30% uplift in client onboarding by deploying KYC AI

Danske Bank expects further job cuts in AI push

Invesco launches behavioural intelligence platform technology

Banks issue urgent warning using AI to shop online raises scam and fraud risks

Panic as advanced AI systems develop a chilling 'secret' language humans are no longer able to decode