When Algorithms Decide Your Credit Score

TL;DR: AI credit scoring has moved beyond your credit card history. Systems now analyze rent payments, utility bills, bank transactions, and even your education history. A Stanford study found AI credit tools are 5-10% less accurate for minorities and lower-income borrowers. The Urban Institute found Black and Brown borrowers are twice as likely to be denied loans. In October 2024, the CFPB fined Apple $25 million and Goldman Sachs $45 million over algorithmic transparency failures. A tenant screening algorithm paid $2.3 million in November 2024 after discriminating against housing voucher users. Starting Fall 2025, your Buy Now Pay Later loans will affect your FICO score. The CFPB says there's "no advanced technology exception" to consumer protection laws, but enforcement is struggling to keep pace with deployment.

What Is AI Credit Scoring?

Traditional credit scores, your FICO score, look at five factors: payment history, amounts owed, length of credit history, new credit, and credit mix. The data comes from the three major credit bureaus: Equifax, Experian, and TransUnion.

AI credit scoring expands that dramatically. New systems use machine learning to analyze hundreds or thousands of variables, drawing on "alternative data" sources that traditional scoring ignores.

What Counts as "Alternative Data"

Data Type	What It Reveals
Rent payments	Whether you pay on time (if reported)
Utility bills	Phone, electricity, internet payment history
Bank transactions	How you spend, save, and manage money
Education	Where you went to school, what you studied
Employment data	Job history, income verification
Device data	How you use devices, browsing patterns
Social media	Your network, posts, and behavior

The Scale of the Problem

The CFPB estimates that 26 million Americans are "credit invisible", no credit history with any major bureau. Another 19 million have credit files too thin or stale to score.

This disproportionately affects Black and Hispanic consumers, recent immigrants, young people, and the recently divorced or widowed.

Alternative data promises to include these people in the credit system. The question is: at what cost?

The Discrimination Problem

AI credit scoring has a fundamental accuracy problem, and it falls hardest on those who can least afford it.

The Stanford Finding

A Stanford preprint study found that AI predictive tools are 5 to 10 percent less accurate for lower-income families and minority borrowers than for higher-income and non-minority groups.

The reason: it's not that the algorithms themselves are explicitly biased. The underlying data is less accurate for predicting creditworthiness in these groups, often because they have limited credit histories or use financial products differently.

The Numbers

Urban Institute (2024): Black and Brown borrowers were more than twice as likely to be denied a loan than white borrowers
UC Berkeley study: African American and Latinx borrowers pay nearly 5 basis points higher in interest rates than credit-equivalent white counterparts, amounting to $450 million in extra interest per year
CFPB examiners: Found disproportionately negative outcomes for Black and Hispanic applicants compared to white applicants in credit card lending

Proxy Discrimination

Credit algorithms are legally prohibited from considering race or ethnicity directly. But they don't need to.

AI systems can pick up on "proxy" variables that correlate strongly with protected characteristics:

Zip code (correlates with race due to residential segregation)
School attended (HBCUs, community colleges, Hispanic-Serving Institutions)
Names on the account
Shopping patterns (stores that serve particular communities)
Social connections

The algorithm doesn't "know" your race. It just knows everything that statistically predicts your race, and uses it.

The Upstart Case: Education as a Weapon

Upstart, a leading AI lending platform, uses education data, including where you went to school, to price consumer loans.

The NAACP Legal Defense Fund and Student Borrower Protection Center raised concerns that this practice discriminates against borrowers who attended HBCUs (Historically Black Colleges and Universities), community colleges, or Hispanic-Serving Institutions.

Their analysis: borrowers from these schools paid more for loans than similarly creditworthy borrowers from other institutions.

Senators Elizabeth Warren, Kamala Harris, and Sherrod Brown demanded answers. Upstart agreed to a voluntary monitorship by civil rights attorneys.

The result? In March 2024, the monitor ended "at an impasse over the appropriate and legally required methodology" for testing discrimination. Translation: they couldn't agree on whether the system was discriminatory because they couldn't agree on how to measure it.

The SEC also investigated Upstart's AI disclosures. In May 2025, they closed the probe, no enforcement action, but no clean bill of health either.

Tenant Screening: Locked Out by Algorithm

The same AI scoring systems are deciding whether you can rent an apartment.

How It Works

Landlords use automated tenant screening services that pull credit reports, eviction records, criminal records, and rental history into a single "score" or recommendation.

A 2024 survey by TechEquity found:

Two-thirds of landlords use algorithmic screening tools
37% rely solely on the algorithm's recommendation
Only 3% of renters knew which company generated their denial

The SafeRent Settlement

In November 2024, SafeRent Solutions, a major tenant screening company, agreed to pay $2.3 million and stop using AI scores to evaluate housing voucher users.

The class action lawsuit, filed in Massachusetts, alleged SafeRent's scoring system disproportionately harmed people using housing vouchers, specifically Black and Hispanic applicants.

The company will no longer use AI-powered scores to evaluate voucher holders and must provide human review of flagged applications.

Racial Disparities in Screening

Research shows stark differences:

Black and Latinx renters are almost half as likely to have rental applications accepted as white respondents (46% and 43% vs. white acceptance rates)
Black and Latino renters submit more applications and incur higher fees compared to white and Asian renters
The combination of financial barriers and limited recourse perpetuates residential segregation

HUD Takes Notice

In May 2024, HUD issued guidance explicitly stating that the Fair Housing Act applies to algorithmic tenant screening. Both housing providers and screening companies have legal responsibility for discriminatory outcomes, regardless of whether a human or algorithm made the decision.

Buy Now Pay Later: The Credit Score Disruption

Starting Fall 2025, your Buy Now Pay Later loans will affect your credit score. This is a major shift.

What's Changing

FICO is introducing two new scoring models, FICO Score 10 BNPL and FICO Score 10 T BNPL, that incorporate BNPL data for the first time.

Affirm began reporting all BNPL loans to Experian as of April 1, 2025. Klarna and others are expected to follow.

How It Affects You

On-time payments could help: Especially if you have limited credit history
Missed payments will hurt: Just like traditional credit cards
Score impact: FICO simulations show most users will see changes of around ±10 points
Multiple loans: Consumers with 5+ Affirm loans typically saw stable or improving scores in testing

The Hidden Risk

BNPL loans are short-term. When you pay one off, you're closing a credit account, which can reduce your average credit age (15% of your FICO score).

Klarna has pushed back, arguing credit bureaus "do not have proper models to responsibly process the data."

Regulatory Context

In May 2024, the CFPB declared BNPL lenders equivalent to credit card providers under the Truth in Lending Act. A January 2025 CFPB study found heavy BNPL users (12+ loans per year) are at greater risk of financial distress.

The Alternative Data Expansion

Credit bureaus are racing to incorporate new data sources.

What's Being Added

Experian Boost: Lets consumers opt-in to have rent, utility, and phone payments considered
Equifax (2023): Expanded mortgage credit reports to include phone, TV, and utility bills
TransUnion research: Found adding rent payments increased scores by an average of 60 points for those included

The Promise

Equifax found alternative data could help score 8.4 million previously unscorable borrowers. For people building credit from scratch, rent and utility data can establish a credit profile in 3-4 months.

By 2023, 62% of financial institutions were using alternative data for credit decisions.

The Risk

Every new data source is a new opportunity for discrimination, errors, and privacy invasion.

The same rent payment that helps you on Experian Boost can hurt you if a landlord reports a disputed late payment. The same utility data can be pulled without your knowledge by lenders who don't use consumer-friendly opt-in systems.

And the more data in the system, the harder it becomes to find and fix errors.

The China Comparison (What's Actually Happening)

Western media often portrays China's "social credit system" as a dystopian nationwide score that controls citizens' lives. The reality is more complex, and the lessons are closer to home than you'd think.

What China Actually Has

China does not have a single nationwide personal credit score. Instead, it has:

Blacklists: Specific penalties for specific misconduct (judgment defaulters, no-fly lists)
Corporate credit: A comprehensive system tracking business compliance, with 80.7 billion records covering ~180 million businesses
Private credit scores: Like Sesame Credit (Ant Group), which is voluntary and functions like a loyalty program

The Relevant Lesson

China's corporate social credit system is extensive and consequential. Companies can be blacklisted, restricted from government contracts, or face public shaming.

But the American system is already using similar logic for individuals: combining disparate data sources, making algorithmic judgments, and creating consequences that ripple across domains (can't get housing because of a credit score that used disputed data).

We don't need to look to China for dystopia. We're building our own version, just with more corporate involvement and less transparency.

Regulatory Response

CFPB Actions

The CFPB has been the most aggressive US regulator on AI credit scoring.

August 2024: "There are no exceptions to federal consumer financial protection laws for new technologies"
October 2024: Fined Apple $25 million and Goldman Sachs $45 million over Apple Card algorithmic transparency failures
2024-2025: Issued supervisory guidance on AI/ML models, requiring lenders to search for "Less Discriminatory Alternatives"

CFPB examiners are now actively testing alternative models when creditors haven't, using open-source debiasing methodologies to identify whether discrimination could be reduced while maintaining accuracy.

Joint Agency Statement

The CFPB, DOJ, EEOC, and FTC issued a joint statement on enforcement against discrimination in automated systems, warning that AI "has the potential to perpetuate unlawful bias, automate unlawful discrimination, and produce other harmful outcomes."

EU Approach

The EU AI Act classifies credit scoring as "high-risk", requiring documentation, testing, and human oversight. This is stricter than current US requirements but doesn't ban the technology.

The Enforcement Gap

Despite regulatory warnings, enforcement has been limited. The Upstart monitorship ended inconclusively. Most discrimination in AI lending is never detected because borrowers don't know they were scored by an algorithm, can't see how the score was calculated, and have limited ability to challenge decisions.

Protecting Yourself

Know Your Data

Free credit reports: AnnualCreditReport.com gives you free reports from all three bureaus
Experian Boost opt-in: Only add data you're confident about, once it's in, errors are hard to remove
Check for errors: 1 in 5 consumers has an error on at least one credit report

Dispute Aggressively

Under the Fair Credit Reporting Act, bureaus must investigate disputes within 30 days
Document everything, disputes sometimes "reappear" after being removed
For complex cases, consider a credit attorney

For Renters

Ask landlords which screening company they use
Request your screening report, you have a right to see it
Pre-screen yourself before applying to identify and fix errors
Know your rights: HUD guidance requires human review of algorithmic decisions

BNPL Strategy

Pay all BNPL loans on time starting now, reporting is coming
Consider whether you want these loans on your credit report at all
Watch for changes to your score after Fall 2025

Alternative Data Awareness

Your rent payments may not be reported unless you opt-in or your landlord reports them
Some services let you add positive rent history, but verify they work with all three bureaus
Be cautious about services that require bank account access to "help" your credit

The Bottom Line

AI credit scoring is expanding what counts as "creditworthiness", from your rent payments to your education to your bank transactions. Proponents say this helps people with thin credit files. Critics say it creates new vectors for discrimination.

The evidence supports both. TransUnion found rent reporting increased some scores by 60 points. Stanford found AI tools are 5-10% less accurate for minorities. SafeRent paid $2.3 million for discriminating against housing voucher users. CFPB fined Apple and Goldman $70 million over algorithmic failures.

The fundamental problem: these systems make decisions using hundreds of variables, and neither borrowers nor regulators can easily see why. When you're denied credit, housing, or charged higher rates, you may never know if an algorithm discriminated against you, because the companies don't have to explain.

The CFPB says there's no technology exception to consumer protection laws. But enforcement lags deployment by years. In the meantime, decisions about your housing, credit, and financial future are being made by systems that perform worse for the people who most need access to credit.

Alternative data could expand financial inclusion. It could also encode existing discrimination into automated systems that operate at scale. Right now, we're getting both, and the people most affected have the least ability to see what's happening.