What if the most important thing that ever happened to your money wasn’t a policy, a rate, or a market crash, but a machine bolted into a wall in north London? On 27 June 1967, a cash-dispensing machine opened at a Barclays branch in Enfield, unveiled to the public by the actor Reg Varney. It didn’t take deposits, it didn’t talk to other banks, and it probably jammed more often than it worked. Yet that machine started a chain reaction that runs straight to the fraud algorithms quietly scanning your card transactions this morning. (Year did cash money take over three answers that actually)
Half a century is a strange amount of time in technology. Long enough for three or four complete rebuilds of how money moves. Short enough that people who queued at that first machine are still alive and still banking. This is the story of how a hole in the wall became a chip, a chip became a phone, a phone became an API, and an API became a model that predicts whether you’ll pay back a loan.
The wall, the card, and the PIN
The first generation of banking technology was almost embarrassingly simple by modern standards. Early machines were standalone cash dispensers. They didn’t verify anything in real time because there was no network to verify against. The breakthrough wasn’t the dispensing mechanism; it was the idea that a machine could stand in for a teller for a single, narrow task.
The United States caught up quickly. Chemical Bank installed the first US ATM at its Rockville Centre, New York branch in September 1969. Like its British cousin, it dispensed cash only. Deposits were still a human job.
Then came the standard that made everything else possible. ISO/IEC 7813, first published in 1985, defined the data format for magnetic-stripe bank cards. That sounds like a footnote. It wasn’t. Once every bank’s card carried data in the same format, machines from different banks could read the same plastic, and shared ATM networks became economically viable. The card-and-PIN model spread because it was boring, standardized, and cheap.
Here’s the practical lesson for anyone building anything today: interoperability beats elegance. The magnetic stripe was a terrible security technology. It was also universal, and universality won for two decades.
What actually went wrong in practice
Anyone old enough to remember the 1990s remembers the fraud. Magnetic stripes are trivially copyable. Skimmers attached to card readers harvested stripe data by the million, and cloned cards drained accounts across borders. Banks spent enormous sums on fraud detection because the underlying card was fundamentally insecure.
The fix arrived in stages. The first EMV chip specification was published in 1996, and EMVCo was founded in 1999 by Europay, Mastercard and Visa to govern it. Chip cards generate a unique cryptogram per transaction, so a copied card can’t simply be replayed. Counterfeit fraud fell sharply in chip-enabled markets. The rollout, though, was a masterclass in how not to manage a transition: merchants hated the terminal upgrade costs, customers hated the extra seconds at checkout, and the US didn’t fully commit for roughly a decade after Europe did.
The mistake people make when reading this history is assuming the better technology wins on merit. It doesn’t. It wins when the incentives line up. Chip cards took off when liability rules shifted fraud costs onto whichever party hadn’t upgraded.
Paper dies slowly, then suddenly
While cards were getting smarter, the back office was still drowning in paper. Every check written in the US physically traveled between banks, sometimes by plane. A storm or a grounded flight could delay settlement for days.
The Check Clearing for the 21st Century Act, known as Check 21, took effect on 28 October 2004. It allowed banks to create substitute checks, essentially image replacement documents, in place of original paper. Suddenly a check could be scanned, transmitted electronically, and cleared without the paper moving at all. The legal fiction was elegant: the image counted as the check, so no bank had to rewrite its rulebook.
Five years later, the branch itself started to dissolve. USAA is widely credited with launching remote deposit capture for consumers via smartphone in 2009, letting customers photograph a check and deposit it from the kitchen table. Treat that as a company claim rather than a settled historical fact, because independent confirmation is thin. But the behavioral shift it represented is undeniable. The camera became a teller window.
This pattern repeats across the whole fifty years. A legal change unlocks a technical one, or a technical change forces a legal one, and then consumer behavior lags behind both by about five years. If you want to know where banking is heading, watch the regulation first.
When the bank became an interface
The next shift was structural rather than mechanical. Banks stopped being places and became data sources.
The EU’s revised Payment Services Directive, PSD2, required member states to transpose it into national law by 13 January 2018. It obliged banks to open payment-account data to licensed third parties through APIs, with the customer’s consent. The UK’s own open banking regime went live on the same day, 13 January 2018, overseen by the Open Banking Implementation Entity, now Open Banking Limited.
The intent was competition. If a startup could read your account data with your permission, it could build a better budgeting app, a smarter lender, or a payment service that undercut the incumbents. The reality was messier. API quality varied wildly between banks, consent flows confused customers, and plenty of consumers decided they didn’t actually want a third party in their account, thank you.
Still, the architectural change was permanent. Once account data could move through a standardized pipe, the bank’s competitive moat shifted from holding deposits to doing something useful with them. That’s the same logic driving the build-versus-buy debates now consuming IT budgets, where teams weigh a cost analysis of building AI in-house against buying it and judging output quality. Banking is discovering that software economics and banking economics are not the same discipline.
The money got faster
Speed was the other frontier. In July 2023 the US Federal Reserve launched FedNow, an instant-payment service that settles transactions in seconds rather than days. Real-time rails already existed in the UK, India, and much of Europe, but FedNow brought the world’s largest economy into the instant-payments era at national scale.
Fast payments change behavior in ways that aren’t obvious. When money arrives instantly, the float disappears. When the float disappears, so does a quiet source of bank revenue and a quiet buffer against fraud. Instant settlement means you can’t claw back a mistaken payment by intercepting it mid-flight. Speed creates its own problems.
That’s a theme worth sitting with. Every efficiency gain in this fifty-year story has been paired with a new failure mode. ATMs created skimming. Chips created card-not-present fraud. Instant payments create irrevocable mistakes. Open banking creates consent fatigue and data exposure. Nobody gets a free lunch. (Alefs 11 year flying car head start meets teslas roadster)
The model in the back office
The current chapter doesn’t have a physical form. Artificial intelligence in banking mostly lives in risk, fraud, and operations, and it has been there longer than most customers realize. Credit scoring has been algorithmic for decades. What changed recently is the sophistication, the data volume, and the regulatory scrutiny.
That scrutiny is now formal. The EU AI Act, Regulation (EU) 2024/1689, was formally adopted in 2024 and sets out a risk-based framework that will apply to AI systems used in credit scoring and other financial services. Its application is phased, running into 2026 and 2027. The practical upshot for banks is that a model which decides who gets a loan is no longer just a business tool; it’s a regulated product with documentation, testing, and explainability requirements.
This is where the fifty-year arc bends back on itself. The 1967 machine made a single binary decision: does this person get cash, yes or no, based on a plastic card and a PIN. The 2026 model makes a probabilistic decision across hundreds of variables. Both are automated judgments about a person’s money. Only one of them has to explain itself to a regulator.
For anyone tracking where this goes next, the broader signal is that adjacent technology curves are converging on the same infrastructure. The same advances in sensors, batteries, and embedded compute that shape the technology trends that will matter by 2030 also determine what a bank branch, a payment terminal, or a wearable wallet looks like. Even something as mundane as choosing between a motion sensor and a presence sensor shows up in branch design and ATM security, where the difference between detecting movement and detecting a person changes how a machine behaves.
A practical timeline you can actually use
| Year | Milestone | Why it mattered |
|---|---|---|
| 1967 | First cash-dispensing ATM, Enfield, London | Proved a machine could replace a teller task |
| 1969 | First US ATM, Rockville Centre, New York | Took the model across the Atlantic |
| 1985 | ISO/IEC 7813 magnetic-stripe standard | Made shared ATM networks possible |
| 1996 | First EMV chip specification | Cut counterfeit card fraud |
| 1999 | EMVCo founded | Gave chip standards a governing body |
| 2004 | Check 21 takes effect (28 October) | Killed the paper check’s physical journey |
| 2009 | Smartphone remote deposit capture | Turned the camera into a teller |
| 2018 | PSD2 transposition and UK open banking (13 January) | Made account data portable |
| 2023 | FedNow launches (July) | Brought instant settlement to the US |
| 2024 | EU AI Act adopted | Put credit models under formal regulation |
Read that table as a sequence of unlocks rather than a list of inventions. Almost nothing on it was the first of its kind. Almost everything on it was the moment a technology became cheap, legal, and universal enough to matter.
Mistakes banks and customers keep making
The most common error in reading banking technology history is assuming adoption is driven by delight. It isn’t. It’s driven by cost, liability, and habit, usually in that order. Chip cards spread when liability shifted. Mobile deposit spread because the alternative was a trip to a branch. Instant payments spread because businesses wanted same-day cash flow, not because consumers demanded speed.
The second error is treating each wave as a replacement. ATMs didn’t kill branches. They changed what branches were for. Apps didn’t kill ATMs. They reduced the number of reasons to visit one. AI won’t kill either; it will quietly absorb the decision-making that happens behind both.
The third error, and the one that costs the most, is underestimating the migration tail. Legacy systems persist for decades because replacing them is risky and unglamorous. There are banks still running core systems designed before the first ATM. Any honest account of this evolution has to admit that the frontier and the back office move at completely different speeds.
For younger readers watching this from the outside, the pattern is worth internalizing early. The people who built fortunes in this industry weren’t the ones who invented the technology. They were the ones who spotted which standard was about to become universal. That instinct is learnable, and it’s the same instinct behind realistic ways to make money as a 12-year-old: find the thing everyone needs, do it reliably, and don’t wait for permission.
Conclusion
Fifty years from a cash dispenser in Enfield to a regulated risk model is a long distance to travel, but the through-line is short. Banking technology has always been about moving a decision closer to the customer and further from the human. The ATM moved cash. The chip moved trust. The API moved data. The model moves judgment.
What comes next will probably look less dramatic than any of it. The next decade of banking technology will likely be about consolidation, regulation, and cleaning up the seams between systems built in five different eras. Less flying car, more plumbing. That’s not a disappointment. Plumbing is what makes the rest possible.
The one thing worth betting on is this: the next genuinely transformative change will arrive with a legal deadline attached, will be adopted reluctantly, and will be described afterward as obvious. That’s how every single one of the last fifty years went.
Frequently asked questions
When was the first ATM invented?
The first cash-dispensing ATM opened at a Barclays branch in Enfield, north London, on 27 June 1967, unveiled by the actor Reg Varney. Earlier cash-dispensing prototypes existed, so “first” depends on whether you mean a standalone cash dispenser or a networked ATM.
When did ATMs arrive in the United States?
Chemical Bank installed the first US ATM at its Rockville Centre, New York branch in September 1969. Like the earliest British machines, it dispensed cash only and did not accept deposits.
What did the EMV chip standard actually change?
The first EMV specification was published in 1996, and EMVCo was founded in 1999 by Europay, Mastercard and Visa to manage it. Chip cards generate a unique cryptogram for each transaction, which makes cloned cards far harder to reuse and sharply reduces counterfeit fraud compared with magnetic stripes.
What is open banking and when did it start?
Open banking requires banks to share payment-account data with licensed third parties through APIs, with the customer’s consent. The EU’s PSD2 had to be transposed into national law by 13 January 2018, and the UK’s own open banking regime went live the same day.
How does the EU AI Act affect banking?
The EU AI Act, Regulation (EU) 2024/1689, was formally adopted in 2024 and uses a risk-based framework that applies to AI systems used in credit scoring and other financial services. Its obligations are being phased in, running into 2026 and 2027, and require documentation, testing, and explainability for higher-risk uses.
Will AI replace bank branches?
Unlikely, and the historical pattern argues against it. ATMs didn’t eliminate branches, they changed what branches do. AI is mostly absorbing back-office judgment like credit decisions and fraud screening, while branches shift toward advice and complex transactions that still benefit from a human present.
