In November 2019, a software engineer named David Heinemeier Hansson posted something on Twitter that went viral within hours. He had just received his new Apple Card. His wife had applied at the same time. They file joint tax returns. They live in a community property state. They have been married for years. His credit limit: 20 times higher than hers.
Then Apple co-founder Steve Wozniak chimed in. The same thing had happened to him and his wife. He got 10 times the limit she did. They have no separate accounts, no separate assets. Nothing about their financial lives is separated.
Neither man could get a straight answer about why. When they called customer service — Apple’s, Goldman Sachs’, whoever would pick up — each representative said the same thing: the decision was made by an algorithm. No one could explain what the algorithm had seen. No one could override it. The black box had spoken, and the black box would not explain itself.
That episode became one of the most visible moments in a story that has been building quietly in financial services for years. And in 2026, what was a controversy is now the operational backbone of how American lenders manage every credit relationship you have.
The Machine That Never Stops Watching
Most people think of credit decisions as a moment in time. You apply for a card, a lender checks your credit, they approve or deny you, and that’s it. The relationship is static until you apply for something else.
That understanding is completely outdated.
Modern credit card companies, mortgage servicers, and personal loan providers have built AI systems that continuously monitor every account they hold. Not periodically. Not once a quarter. Continuously — updating risk scores in real time, analyzing every transaction, every payment, every behavioral signal your account generates.
Here is a partial list of what these systems watch. How much you spend each month relative to your usual patterns. Where you shop — and whether those categories are changing. Whether you are paying your full balance or shifting toward minimums. How close you are to your credit limit. Whether you recently opened new accounts. Whether you made a cash advance, which most AI risk models treat as a serious stress signal. Whether your payment timing is shifting — paying on day 25 instead of day 5, for example. Whether you are buying things that statistically correlate with financial stress: pawn shop transactions, overdraft fees at your bank, divorce attorney consultations, medical bill payments showing up repeatedly.
Synchrony Financial, one of the largest store card issuers in the United States, has publicly discussed using AI to understand «normal customer behavior and flagging deviations.» Mastercard has described systems capable of simulating «things that haven’t happened yet but might happen in the future.» Citibank, according to its own technology reports, achieved a tenfold increase in fraud detection capability through AI while simultaneously reducing false positives tenfold — meaning the system got dramatically better at identifying patterns without mistakenly flagging normal behavior.
What nobody emphasizes in these press releases is that the same behavioral modeling used to detect fraud is also used to manage risk on every account — including yours.
When the Algorithm Acts First
The most consequential thing these systems do is not detect fraud. It is decide, in advance, how much credit to extend to you — and whether to pull it back.
A man we will call Mr. M held an American Express card with a £30,000 credit limit. Between October and December 2023, he made several transactions with a particular merchant. He discussed those transactions with American Express in a phone call on December 27, 2023. The next day — the very next day — his credit limit was cut from £30,000 to £16,700, and his interest rate was increased. American Express told him the system had «flagged» his account due to patterns it associated with that merchant. When he asked for a detailed explanation of what specifically had triggered the decision, the company told him it was «not at liberty to discuss the details of its assessment.» The case eventually reached the UK Financial Ombudsman, who documented the entire sequence. Mr. M received £100 as compensation for the inconsistent information he was given during his attempts to understand what had happened.
Another documented case involves a man — call him Mr. H — who held a credit card with Creation Financial Services. He had been a customer for years. He always paid his balance in full. On May 8, 2024, Creation slashed his credit limit from £7,000 to £800 without any prior communication. Three days later, he tried to pay for petrol and his card was declined at the pump. He did not know his limit had been changed. He called Creation. They acknowledged they had sent him a letter — which he had not yet received — and told him the decision was based on «information received from credit reference agencies.» The UK Financial Ombudsman documented this case too. Mr. H’s complaint was not upheld. Creation’s criteria, the investigator found, were applied correctly. The fact that a loyal customer with a perfect payment history had his limit cut by 89% with no real explanation and found out at a petrol station was, apparently, within the rules.
These are not isolated incidents. A 2025 PYMNTS analysis found that one-third of consumers who experienced a credit limit adjustment in the past three years felt at least one decision was unfair. Among subprime borrowers, that figure climbs to nearly 60%.
The Predictive Layer: Seeing Your Crisis Before You Do
Here is where the story gets genuinely unsettling — and where most consumer reporting falls short.
AI delinquency prediction systems do not wait for you to miss a payment. They are designed to identify what researchers call «silent struggling borrowers» — people who are still paying on time but whose behavioral patterns have shifted in ways that statistically predict a future default. These systems, according to research published in January 2026, can flag financial stress 30 to 60 days before a borrower misses their first payment.
Think about what that means in practice. You lose your job on a Monday. You have not missed a single payment. You have not called your bank. You have not changed anything on paper. But in the week after your job loss, your spending patterns subtly shift. You stop eating out. You cancel a streaming subscription. You make one small cash withdrawal. You pay your credit card statement on day 27 instead of day 5. None of these things individually signals distress. Together, to an AI model trained on millions of accounts, they tell a story.
The lender’s system sees that story. It quietly adjusts your risk tier upward. In the background, it begins deciding whether to reduce your credit line, freeze future credit increases, or flag your account for closer monitoring. You receive no notification. Nothing in your relationship with the bank visibly changes. But the terms under which the bank views you have already shifted.
And here is the part that stings: this happens precisely at the moment you might need your credit line most. The month you lose your job is the month your safety net quietly gets smaller, because the algorithm decided to act before you did.
The Apple Card Aftermath: What Actually Changed
The Apple Card controversy of 2019 triggered a formal investigation by the New York Department of Financial Services. The investigation ran for two years. In March 2021, the NYDFS released its findings: Goldman Sachs did not intentionally discriminate against women. Applications from women and men with similar credit characteristics generally had similar outcomes.
But the regulator’s own superintendent, Linda Lacewell, added a caveat that received far less attention than the headline clearance. She said the case should serve as a reminder that credit-scoring models and anti-discrimination laws need a refresh, almost fifty years after the Equal Credit Opportunity Act was written. The problem, she was saying, is not that Goldman acted with intent. The problem is that the legal framework for catching unintentional discrimination produced by algorithms was not adequate.
That gap has not closed. In October 2024, the CFPB fined Apple $25 million and Goldman Sachs $45 million for separate failures related to the Apple Card — this time involving disputed charges and operational breakdowns rather than credit limits. The algorithm question was never fully resolved. The legal standard for what «fair» means when a machine makes financial decisions affecting millions of people remains contested and incomplete in 2026.
What Lenders Are Not Required to Tell You — But Should
Under the Equal Credit Opportunity Act, lenders are required to provide a specific, written adverse action notice when they deny credit, reduce a credit limit, or change your terms unfavorably based on information from a credit report. The notice must state specific reasons.
In practice, the CFPB’s January 2025 Supervisory Highlights found that multiple institutions were using generic, checklist-style adverse action notices that failed to provide specific individual explanations — particularly when an AI model had made the decision. The CFPB was explicit: the fact that an algorithm produced the decision does not reduce the legal obligation to explain it specifically. That finding was a warning shot to the industry, not a resolution.
The regulation requiring explainability is there. The enforcement of it — particularly at the federal level under the current administration — has weakened. State-level protections vary dramatically. Colorado’s SB 24-205, effective in 2026, requires auditability and transparency in high-risk AI systems, including credit decisions. New Jersey, Massachusetts, California, and New York all maintain broader anti-discrimination frameworks that cover algorithmic lending practices. If you live in one of these states, your legal recourse when an AI makes a credit decision you cannot understand is meaningfully stronger.
How to Protect Yourself Right Now
Understanding that AI systems are continuously monitoring your financial behavior changes how you should manage your credit — not out of paranoia, but out of the same practical awareness that leads you to drive carefully near a speed camera.
The most important thing you can do is make your financial behavior boring and consistent. AI risk models flag anomalies. A sudden surge in spending, a category change, a shift in payment timing — all of these generate signals. Pay your balance or as much of it as possible on the same day each month. Keep credit utilization below 30%. Avoid cash advances at all costs: most AI models weight them heavily as stress indicators. These are good financial habits anyway; in the age of AI credit surveillance, they are also a form of self-protection.
Monitor your own accounts proactively. Tools like Experian’s credit monitoring, Credit Karma, and your bank’s own transaction alerts are free and give you visibility into what the system is seeing. If your utilization spikes unexpectedly — because a limit was quietly reduced — you want to know before it affects your score further.
Pull your full credit reports from all three bureaus — Experian, Equifax, and TransUnion — at least twice a year. Dispute every inaccuracy. Under the Fair Credit Reporting Act, bureaus have 30 days to investigate. An error on your report that feeds into an AI risk model can cascade into limit reductions, rate increases, and loan denials that are entirely preventable.
If you receive an adverse action notice and the reason stated is vague, push back. You are entitled to specific reasons. Call the lender, document the call, and ask directly which data points produced the decision. In many states, you have explicit rights to this information. If the response is unsatisfactory and you believe the decision reflects bias or inaccuracy, file a complaint with the CFPB at consumerfinance.gov and with your state attorney general’s office.
Finally — and this is perhaps the most empowering shift in perspective available to you — use AI yourself. Free tools like Trim, Cleo, YNAB, and NerdWallet’s cash flow analyzers use the same behavioral pattern analysis that lenders use, but in your favor. They show you when your spending is trending toward stress before the bank’s algorithm sees it. They help you make the correction before the correction is made for you, quietly and without warning, the way Mr. H found out at the petrol station.
The surveillance is real. The asymmetry of information is real. But so is your ability to understand the system and use that understanding to stay ahead of it.
Disclaimer: This article is for informational purposes only and does not constitute legal or financial advice. Consumer protection laws, AI regulation, and lender practices vary by state and jurisdiction and are subject to change. Real cases cited are drawn from publicly available regulatory documents and verified reporting. Always consult a qualified financial advisor and, where appropriate, a licensed attorney if you believe your rights under the Equal Credit Opportunity Act or the Fair Credit Reporting Act have been violated.