New York's financial regulator investigated the Apple Card algorithm for gender bias and found no unlawful discrimination

Verification status: CHECKING — handed to the checker, not yet graduated. NOT verified, NOT green. This is a financial-regulator record about an algorithmic-underwriting bias allegation, an honest-negative, not a client testimonial. No green badge is sought; the war-room never sets verified.

The problem

The Apple Card was one of the first consumer credit products whose underwriting was widely alleged in public to be biased by a machine-learning model. “In November 2019, allegations surfaced on Twitter regarding discrimination by the Bank in extending credit for Apple Card” (source). The complaint that set it off came from a tech entrepreneur who reported that, despite filing joint tax returns and living in a community-property state, “he was offered a credit limit on an Apple Card 20 times higher than her offer. In a series of tweets, beginning November 7, 2019, the Consumer ascribed the difference in credit limits to sex discrimination” (source). TechCrunch, reporting on the eventual findings, described the same trigger: the entrepreneur “received a credit limit that was 20 times higher than what his wife was offered - even though the couple filed joint tax returns and his wife had a higher credit score than he did” (source). The allegation, amplified by a co-founder of Apple who said he saw the same pattern with his own wife, made the Apple Card a test case for whether an algorithm was encoding sex-based bias into credit decisions (source).

What was built

Apple and Goldman Sachs launched the Apple Card in August 2019, with Goldman Sachs, “a New York State-chartered bank,” responsible for “the Apple Card credit policy and underwriting decisions” (source). Because Goldman Sachs is New York-chartered, its Apple Card underwriting fell within the Department’s fair-lending jurisdiction, and “the Department, authorized to enforce federal and state fair lending laws and mandated to protect consumers, launched an investigation into the allegations” (source). The Department’s Consumer Examinations Unit, “which specializes in statistical analysis of lending outcomes to detect disparate impact,” tested the underwriting model itself, not just the anecdotes (source). Its review covered a large population: the Department’s press release states the finding rested on “analysis of underwriting data for approximately 400,000 New York State applicants for the Apple Card” (source).

The outcome

The regulator did not substantiate the bias claim. “DFS reviewed the Apple Card policies and practices for evidence of both disparate treatment and disparate impact and found that women and men with equivalent credit characteristics had similar Apple Card application outcomes” (source). On intentional bias, “the Department did not find evidence of unlawful intentional discrimination against women or members of other protected classes,” and “likewise, the Department did not find evidence of disparate impact fair lending violations in the Bank’s Apple Card credit lending decisions” (source). The Department summarised the whole inquiry the same way, stating that its review “did not produce evidence of unlawful discrimination against applicants under fair lending law” (source). TechCrunch reported the bottom line identically: “The Department concluded that there was no ‘unlawful discrimination’ against applicants under fair lending law” (source). Superintendent Linda A. Lacewell framed the result as a clearance, not an endorsement: “While we found no fair lending violations, our inquiry stands as a reminder of disparities in access to credit that continue nearly 50 years after the passage of the Equal Credit Opportunity Act (ECOA)” (source).

Where the Department did fault the program was service and transparency, not lending. Its review “did not produce evidence of deliberate or disparate impact discrimination but showed deficiencies in customer service and transparency, which the Bank and Apple have since taken steps to remedy” (source). Those remedies were concrete: “the Bank, upon re-review of both sets of credit files, raised the credit limits of both women to match the credit limits of their respective spouses within days after their complaints surfaced. At that time, the Bank also eliminated the six-month waiting period for appeals of credit decisions” (source), and the press release records that the firms “modified a policy that previously required approved applicants to wait six months before appealing credit terms” (source). TechCrunch, reporting independently, recorded the same two remedies: “Goldman Sachs re-reviewed the credit files of the some of the women who had been initially been offered dramatically lower credit scores than their spouses, and decided to raise their limits to match those of their spouses,” and “at the time, the bank also eliminated the six-month waiting period for appeals on credit decisions” (source).

Weakest load-bearing source. The load-bearing evidence is Tier 1 and primary: the DFS report and its companion press release, from which every finding and figure above is quoted verbatim. Their honest limit is that they are the regulator’s own account of its investigation, not an adversarial court judgment, and the underlying regression workpapers and the Bank’s model documentation are not public, so the “no disparate impact” conclusion cannot be independently re-run from the record. The one Tier-2 source, TechCrunch, is an independent newsroom used to corroborate the headline finding, the trigger and the two remedial actions; note that its line about the wife’s “higher credit score” restates the original allegation, which the Department’s analysis of ~400,000 applicants did not bear out, so it is presented here as what was claimed, not as a verified fact.

How this was verified. Method: every finding and figure is quoted verbatim from two Tier-1 New York State Department of Financial Services primaries: the “Report on Apple Card Investigation” (March 2021; the report PDF, retrieved this session and archived to the Wayback Machine at web.archive.org/web/20260903102748, with the cleaned full text saved to sources/nydfs-apple-card-report-2021-03-flattext.txt) and the DFS press release “DFS Issues Findings on the Apple Card and Its Underwriter Goldman Sachs Bank” (23 March 2021; archived Wayback 20260529003349, saved to sources/), which carries the ~400,000-applicant figure and the Superintendent’s quote. Independent corroboration is TechCrunch (Tier 2; archived Wayback 20260114190958). This is an honest-negative: the regulator found no unlawful discrimination. No confirmation was sought from Goldman Sachs or Apple; the badge never depends on a subject confirming its own conduct. Verified 2026-09-03 (checking round).

Sources

1. New York State Department of Financial Services · Report on Apple Card Investigation · March 2021 · https://www.dfs.ny.gov/system/files/documents/2021/03/rpt_202103_apple_card_investigation.pdf — Tier 1 (primary; the investigating regulator’s own report finding no unlawful discrimination; landing page https://www.dfs.ny.gov/reports_and_publications/202103_report_apple_card_investigation; archived Wayback 20260903102748; cleaned full text saved to sources/nydfs-apple-card-report-2021-03-flattext.txt).
2. New York State Department of Financial Services · DFS Issues Findings on the Apple Card and Its Underwriter Goldman Sachs Bank (press release) · 23 March 2021 · https://www.dfs.ny.gov/reports_and_publications/press_releases/pr202103231 — Tier 1 (primary; carries the “approximately 400,000 New York State applicants” analysis figure, the “no unlawful discrimination” conclusion, the six-month appeal-policy change, and Superintendent Lacewell’s quote; archived Wayback 20260529003349; saved to sources/nydfs-press-release-pr202103231-2021-03-23.html).
3. TechCrunch · New York’s Department of Financial Services says Apple Card program didn’t violate fair lending laws · 23 March 2021 · https://techcrunch.com/2021/03/23/new-yorks-department-of-financial-services-says-apple-card-program-didnt-violate-fair-lending-laws/ — Tier 2 (independent newsroom; corroborates the no-discrimination finding and the November 2019 trigger and 20x disparity; archived Wayback 20260114190958; quotes saved to sources/techcrunch-2021-03-23-quotes.txt).

Related case files

• Meta’s algorithmic ad-delivery system and the DOJ Fair Housing Act settlement — the mirror image outcome: a US enforcement action that DID find an ad-targeting algorithm unlawful, where this Apple Card inquiry cleared the underwriting model.
• Louis v. SafeRent and the $2.275m algorithmic tenant-screening settlement — another consumer-credit-adjacent scoring algorithm challenged for discrimination, contrasting a settlement of liability with this finding of none.
• Manulife’s MAUDE AI underwriting engine — another AI underwriting model in financial services, on the opposite side of the ledger: a self-reported efficiency outcome rather than a regulator’s fair-lending adjudication of the same kind of model.