There is a postcode in east London — one of dozens you could pick, really — that sits at the border of several things simultaneously: a neighbourhood with historically high ethnic minority population density, a rising property market, and several decades of lending data that reflects credit decisions made under conditions that are no longer legal and probably were not then. That data is not marked as discriminatory. It is simply logged as outcomes — accepted applications, rejected applications, loan-to-value ratios, default rates — and it feeds, in aggregated form, into the models that some UK mortgage lenders are now using to process new applications. The algorithm doesn’t know about the history. It just knows about the patterns. That distinction, the EHRC has warned, is precisely where the Equality Act problem begins.
The Equality and Human Rights Commission, Britain’s independent equality watchdog, has issued a clear warning: AI systems being used in mortgage and home loan decisions risk breaching the Equality Act 2010 by producing discriminatory outcomes, even when no discriminatory intent is present. The Act prohibits both direct and indirect discrimination against individuals on the basis of protected characteristics — race, sex, disability, age, and others. Crucially, the prohibition on indirect discrimination applies to practices that appear neutral but produce disproportionately disadvantageous outcomes for a protected group. An algorithm that was not designed to discriminate against any particular group can still meet that legal threshold if its outputs consistently disadvantage one. The EHRC’s message to financial institutions is that ignorance of this dynamic does not constitute a legal defence.
| Category | Details |
|---|---|
| Watchdog Body | Equality and Human Rights Commission (EHRC) — UK’s independent equality and human rights watchdog; legal powers to investigate bodies suspected of breaching equality law |
| Legal Framework | Equality Act 2010 — prohibits both direct and indirect discrimination based on protected characteristics including race, sex, disability, and age; applies to automated decisions as much as human ones |
| Core Warning | AI systems trained on biased or incomplete historical data may produce mortgage and credit outcomes that disadvantage applicants on the basis of protected characteristics — even without explicit discriminatory intent |
| Mechanism of Bias | “Proxy discrimination” — algorithms using variables such as postcode, spending patterns, or social media data that correlate with protected characteristics (race, income, location) without explicitly referencing them |
| Related Parliamentary Action | February 4, 2026 — Joint Committee on Human Rights examined AI regulation; EHRC chair Mary-Ann Stephenson testified that resource constraints are the greatest hurdle to effective AI oversight |
| FCA Parallel Position | Financial Conduct Authority requires fairness and explainability in consumer credit AI; MCOB rules require lenders to treat customers fairly and communicate clearly — but no specific AI disclosure rule exists yet |
| EHRC Monitoring Programme | Working with 30+ local authorities and public bodies to audit AI use in essential services; separate programme exploring algorithmic discrimination in online recruitment and facial recognition |
| EHRC Recommendation | Organisations must understand their AI systems’ impact, maintain human oversight, ensure transparency with applicants, and test for discriminatory outcomes before and after deployment |
The mechanism that generates this risk has a specific name in the regulatory literature: proxy discrimination. An algorithm processing a mortgage application does not ask for a person’s ethnicity or national origin. It processes postcode. It processes spending pattern data. It processes employment sector, device type, social media presence. Each of these variables is technically race-neutral. But in practice, across a sufficiently large dataset, they correlate with race, with income bracket, with the specific neighbourhoods where certain communities have historically been concentrated. The model, optimizing for a definition of risk that is built on historical outcomes, learns those correlations without being taught them. It then applies them to new applicants in ways that may disadvantage people from those communities at rates that exceed their actual credit risk — and that, if challenged, would likely qualify as indirect discrimination under the 2010 Act. The 42BR Barristers’ legal analysis, reviewing the current UK case landscape, noted that while discrimination arguments rarely appear in mortgage proceedings, they have already reached senior courts, and the trajectory is clearly pointing toward more of them.
The EHRC’s regulatory capacity to act on these concerns is itself a live issue. On February 4th, 2026, the Joint Committee on Human Rights took oral evidence from three key regulators — the EHRC, the Information Commissioner’s Office, and Ofcom — on whether they have the resources and powers to address AI-related human rights harms. EHRC chair Mary-Ann Stephenson was direct: resources were the greatest hurdle. There are at least thirteen regulators in the UK with some AI-related remit. None is dedicated exclusively to AI oversight. The EHRC has legal powers to investigate bodies suspected of breaking equality law but has warned of the financial imbalance between regulators and the technology industry — a gap that, in practice, constrains the regulator’s ability to take on complex, resource-intensive investigations involving proprietary algorithmic systems.
The FCA’s position adds a parallel layer of concern. Under the Mortgage Conduct of Business rules, lenders are required to treat customers fairly and ensure communications are clear and not misleading. The general fairness principles that run through FCA regulation apply, in theory, to AI-driven decisions. But as 42BR’s analysis observed, there is currently no specific FCA rule requiring lenders to disclose that an AI system is involved in calculating loan terms or making eligibility decisions. A borrower can receive a mortgage offer generated partly or entirely by an algorithm whose logic they cannot interrogate and whose biases are not independently audited — and the current regulatory framework does not necessarily require that to change before a harm occurs.
Watching this regulatory gap widen against the backdrop of accelerating AI adoption in financial services, there’s a feeling that the enforcement moment — the case that makes the legal theory concrete — is coming closer rather than remaining theoretical. The EHRC has recommended that organisations understand their AI systems’ impact before deployment, conduct ongoing bias testing, maintain meaningful human oversight, and ensure transparency with applicants about how decisions are made. Those are reasonable requirements. They are also, at the moment, largely voluntary. The distance between where the watchdog says organisations must get to and where most organisations currently are is the space in which the next discriminatory mortgage decision is probably already being made.
