The broom in the room
In Nexus, Yuval Noah Harari makes an argument that ought to be uncomfortable for anyone who works in financial services. Information networks, he contends, have never optimised for truth. They optimise for cohesion — for the flow that holds the network together. Truth is expensive and inconvenient. Flow is cheap and it keeps everyone in the room.
Which is why, in Harari's telling, more information has never automatically produced more wisdom. We have more data than any civilisation in history and we are still cooking the planet. The stone tablet, the book, the printing press, the internet — each expanded distribution, and each left the harder question of what any of it was worth to whoever happened to be holding it.
Banking is one of the oldest information networks we have, and it is not exempt. A bank is not a vault. It is a set of claims about who somebody is, what they earn and what they can bear. And those claims have never been more abundant. We hold more data on a borrower today than a branch manager in 1985 accumulated over an entire career.
It has not made lending wiser. It has made it faster, and occasionally spectacularly wrong — because the network never optimised for truth either. It optimised for throughput. More applications processed, more documents uploaded, more decisions rendered per hour. Volume got mistaken for knowledge.
The enchanted broom in lending was never the AI. It was the process.
Harari borrows the Sorcerer's Apprentice to describe what happens when we animate a system we cannot then command. In origination, the broom is old and we built it ourselves. Upload a document, extract the numbers, score the applicant, render the decision. Automate it, incentivise it, scale it, and it will keep fetching water long after the room has flooded.
Nobody at the Commonwealth Bank of Australia chose to write a billion dollars of fraudulent home loans. A network optimised for throughput did precisely what it was built to do — faithfully, at speed, against documents that had quietly stopped meaning anything.
AI did not break the process. AI gave the broom infinite buckets.
So the answer is not another layer of detection bolted onto the same rail: a smarter model to catch a smarter forgery, forever. That is an arms race, and the forgers are better capitalised. The answer is to change what the network runs on. Not to verify the document harder — to stop needing the document at all.
That is what open banking is. There are three arguments for it. Two of them are about growth. The third is about survival.
One: speed wins the deal
Lending is a race. The correlation between response time and conversion is one of the most durable findings in consumer credit, and it is not subtle.
Customers shopping for a loan are not waiting patiently. They are applying to three lenders on a Sunday evening and taking the first credible yes. The lender who comes back in minutes wins. The lender who comes back in days is doing free credit assessment work for a competitor.
Speed in origination is almost entirely a data problem. Underwriting needs income, expenses, liabilities and account conduct. Right now, most institutions get that by asking the customer to export PDF statements from another bank's app, upload them, and then running OCR or screen-scraping over the result. That is a multi-day round trip with a human in the middle and a failure rate at every step.
Open banking collapses that. The customer consents once, and the data arrives structured, categorised and complete. Not a scan of a document that represents the truth. The truth.
Two: aggregation keeps them on your property
Nobody has one bank anymore. They have a transaction account here, a mortgage there, a buy-now-pay-later facility somewhere else, super in a fourth place and crypto in a fifth. Gen Z are the extreme case — they will hold five financial relationships before they hold one meaningful balance. Loyalty, in the old sense, is gone.
What replaces it is the app they open. Whoever holds the aggregated view of the customer's financial life holds the relationship, regardless of who holds the deposits. Open banking is the mechanism by which a challenger, a mutual or a community bank can pull in balances, transactions and assets from every other institution and present the whole picture inside their own experience.
You cannot coach a customer whose financial life you can only see a tenth of. Aggregation is not a feature. It is the precondition for any bank that intends to help rather than extract.
Three: you can no longer trust a document
And then there is the one that matters most — the survival argument.
In February 2026, the Commonwealth Bank of Australia reported itself to police and the corporate regulator over approximately A$1 billion in home loans suspected of having been obtained fraudulently — including through AI-generated documents.
The bank discovered the suspected fraud partly through two whistleblowers, and moved after rival NAB was allegedly defrauded of around $150 million. Complaints lodged through CBA's internal reporting platform accused a mortgage broker and a lender in the private banking division of forging income statements. The fraud was not caught by detection systems. It was caught because two people pressed a button.
Get the sequence right, because it is the whole point. CBA did not knowingly lend against AI-generated paper. CBA lent against paper it had no way of knowing was AI-generated. Fabricated payslips. Draft tax returns never lodged with the ATO. Shell companies that appeared to have traded for years. Deposits routed to disguise their origin. Documents that were internally consistent, professionally formatted, and entirely false.
ASIC Chair Joe Longo has told a parliamentary committee that AI-assisted document fraud looks like a real emerging issue for the banks. Westpac and ANZ are also in scope. This is not a CBA story. This is the story of every lender still making billion-dollar decisions on the basis of a PDF.
The wider collapse of analogue proof
Content warning
If you think document fraud is a niche financial-crime problem, look at what is happening to identity generally.
In the UK, the National Crime Agency and the Internet Watch Foundation have just issued new guidance to parents because criminals are scraping ordinary, publicly posted photographs of children and running them through AI to manufacture abuse material. A criminal gang took pupils' images from a school website, generated more than 100 sexual images of those children, and attempted to blackmail the school into paying to keep them offline.
It is the worst thing in the world, and it is also a data point. A photograph of a child at football is now sufficient raw material for a convincing forgery of that child. Sit with what that implies about a payslip.
The voice channel has already gone the same way. Generative AI can clone a voice from as little as three seconds of audio scraped off social media, and caller ID spoofing means the number on the screen is fake too. More than one in ten banks has lost over $1 million to deepfake voice fraud. A finance worker in Hong Kong was induced to pay out $25 million after attending a video conference in which the CFO and several colleagues were all deepfakes.
Audio has fallen. Video is falling. The idea that a bank can establish identity by looking at a document, listening to a voice, or watching a face on a screen is finished. It was the foundation of KYC for a century and it is now a liability.
What open banking actually is
Everyone sells open banking as a convenience feature. It is not. It is a provenance system.
When a customer consents to share their data through open banking, they authenticate directly with their own institution, inside that institution's security perimeter, and the data travels bank-to-bank over an accredited channel. There is no document to forge. There is no broker in the middle retyping a number. There is no scan, no OCR, no upload, no metadata to scrub. The transaction history is the transaction history because the bank that holds it says so.
A PDF asserts a fact. An open banking feed is the fact.
Which means the three arguments are actually one argument. The same rail that makes decisioning fast is the rail that makes decisioning safe, and the same rail that lets you aggregate a customer's whole financial life is the rail that lets you verify it. You do not have to trade speed against integrity. That trade-off only exists in the analogue world — and it is the reason lenders have historically been slow. Every day of delay was a day spent squinting at paperwork, trying to decide whether to believe it.
Open banking removes the squinting.