A twelve-page comment letter submitted on August 14, 2026 by the National Fair Housing Alliance (NFHA) to Ranking Member Maxine Waters of the U.S. House Financial Services Committee. It responds to the Request for Information on AI Risks and Modernization in Financial Services issued by committee Democrats on July 7, 2026 (Source: democrats-financialservices.house.gov). NFHA answers five of the RFI's numbered questions — on explainability standards, pre- and post-deployment controls, oversight during reduced agency enforcement, liability allocation and safe harbors, and supervision of property technology — and appends a bibliography of its prior AI-related filings, papers and testimony (Source: nationalfairhousing.org).
The letter identifies NFHA's Responsible AI Lab (RAIL) as the internal program from which much of its AI work originates.
Explainability standards (Question 10)
NFHA argues for a consumer right to explanations for automated decisions and for required testing that clarifies the reasoning and design behind AI systems, describing explainability standards as safeguards rather than constraints. Its concrete proposals are three:
- Build on the Equal Credit Opportunity Act. Regulation B requires creditors to give adverse-action notices disclosing the principal reasons for a denial, and those reasons must relate to and accurately describe the factors the creditor considered (15 U.S.C. 1691(d); 12 CFR 1002.9). The letter states that the complications raised by AI systems do not relieve creditors of that obligation, and that models must reliably produce consistent, specific reasons a consumer can understand.
- Extend comparable requirements beyond ECOA. For AI systems used in financial services and housing that ECOA does not reach, NFHA asks for comparable legal requirements producing reasons that are both actionable and accurate.
- Publish or make inspectable the variable lists. A public list of the variables an AI system considers in approving or disapproving a consumer, or a list maintained for regulatory inspection, is described as forcing developers to address fairness before release, enabling regulators to identify systems for further scrutiny, and allowing public evaluation of whether particular variables are appropriate.
The letter notes that the Consumer Financial Protection Bureau withdrew Consumer Financial Protection Circular 2022-03, on adverse-action notification in connection with credit decisions based on complex algorithms, 87 FR 35864 (June 14, 2022), on May 12, 2025 (Source: federalregister.gov).
Pre- and post-deployment controls (Question 11)
NFHA states that AI systems must comply with civil rights, fair housing and consumer protection law across the full lifecycle — design, model buildout, deployment, monitoring and updates — including the prohibitions on disparate treatment, disparate impact and proxy discrimination.
It points Congress and regulators to its own Purpose, Process, and Monitoring (PPM) framework for auditing algorithmic systems, published February 17, 2022 (Source: nationalfairhousing.org). Five elements are enumerated:
- the selected model is the least discriminatory alternative (LDA) meeting the design objective, in compliance with the Fair Housing Act, ECOA, and 12 U.S.C. 4545 in the case of Fannie Mae and Freddie Mac;
- any fairness constraint applied at the data, model or post-modeling stage is evaluated at the model-assessment stage;
- known limitations and assumptions of the model are documented and reviewed;
- circumstances in which the model may or may not be used outside its intended scope are documented and reviewed;
- differences in patterns between the training and serving stages are monitored, and can inform when and how a model is retrained.
The letter calls independent third-party audits and impact assessments, both pre- and post-deployment, essential in high-stakes areas such as housing and financial services.
Oversight amid reduced enforcement (Question 20)
The RFI question asks what Congress can do to maintain oversight of third-party vendors during a period of reduced agency enforcement. NFHA's answer has four strands.
Against exclusive regulatory authority. No single regulator should hold exclusive authority over AI systems in housing and financial services, on the ground that exclusive oversight invites regulatory capture and produces harm through inaction.
Against federal preemption of state law. The letter treats preemption as the central risk, drawing an analogy to the pre-2008 period in which federal banking agencies used preemption authority to displace state fair-lending and consumer-protection oversight. It cites Cuomo v. Clearing House Assn., L.L.C., 557 U.S. 519 (2009), and the Dodd-Frank Act's limits on preemption by federal banking regulators. It then characterizes current federal action as an attempt to prohibit states from setting AI guardrails, citing Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence," 90 FR 58499 (Dec. 11, 2025), and the Federal Trade Commission's proposed policy statement on the suppression of accuracy in AI systems, 91 FR 41638 (July 7, 2026). NFHA's July 31, 2026 comment on that FTC proposal argued that the Commission had failed to make the factual and legal case for deception under the FTC Act and relied on unreliable data; that consumers expect and support anti-bias guardrails; that the preemption case is legally deficient; and that state fair housing and fair lending laws have coexisted with Section 5 of the FTC Act for decades (Source: regulations.gov).
For private enforcement. Individuals hold private rights of action under the Fair Housing Act and ECOA (42 U.S.C. 3613; 15 U.S.C. 1691e), but the letter argues that discrimination by AI systems is difficult for an individual consumer to detect. It presses the role of qualified fair housing enforcement organizations (24 CFR 125.103) funded through the Fair Housing Initiatives Program (42 U.S.C. 3616a), quoting Congress's finding on making FHIP permanent that "the proven efficacy of private nonprofit fair housing enforcement organizations and community-based efforts makes support for these organizations a necessary component of the fair housing enforcement system" (Pub. L. No. 102-550, § 905, 106 Stat. 3868–3872). It states that the administration has continued a pattern of actions to defund these organizations, citing an August 6, 2026 letter from Senators Booker, Van Hollen and Warnock to HUD Secretary Scott Turner (Source: booker.senate.gov).
Evidence from enforcement practice. Two of NFHA's own matters are offered as demonstrations that private enforcers fill gaps left by regulators: the 2019 settlement with Facebook over its advertising platform (Source: nationalfairhousing.org), and a May 2024 complaint with the Fair Housing Rights Center in Southeastern Pennsylvania and the Housing Equality Center of Pennsylvania against a national tenant-screening software company over restrictions on housing choice voucher holders, settled July 1, 2026 with corrections to the policies at issue (Source: nationalfairhousing.org).
The letter also cites a 2026 FAccT paper finding that federal regulator expectations and supervision underpinned much of financial firms' prior fair-lending compliance work, and argues that without that incentive the work is at risk (Source: doi.org).
Liability and safe harbors (Question 22)
This is the answer that drew press coverage. NFHA states that third-party AI systems can drive discrimination and that there is no basis for immunizing them from existing fair housing and fair lending laws or from future AI-specific laws.
The affirmative formulation is role-differentiated rather than uniform: civil rights, fair housing and consumer protection laws "must apply to every actor in the AI ecosystem, with clear legal responsibilities and liability allocation based upon roles," naming developers, deployers, third-party vendors and users. To make that allocation workable, the letter asks policymakers to require accessible records explaining what an AI system was trained on, how it was built and how it is monitored. It adds that existing federal law — the Fair Housing Act, ECOA, and 12 U.S.C. 4545 for Fannie Mae and Freddie Mac — already applies to AI systems, and that many traditional rules of liability allocation remain applicable.
PropTech supervision (Question 34)
NFHA answers by enumerating what has been rescinded. It cites the Government Accountability Office's September 2025 report on property technology for homebuying, GAO-25-107201, and states that GAO's recommendation that the FHFA Director give Fannie Mae and Freddie Mac written direction on fair-lending compliance and supervision is one FHFA neither agreed nor disagreed with, and that it remained open as of the letter's date (Source: gao.gov).
Nine FHFA instruments are listed as rescinded or repealed: Orders 2025-OR-B-1, 2025-OR-FHLMC-1 and 2025-OR-FNMA-1; Orders 2021-OR-FHLMC-2 and 2021-OR-FNMA-2; Advisory Bulletins AB 2024-07, AB 2024-06 and AB 2023-05; and 12 CFR part 1293, removed by the rule at 91 FR 5278 (Feb. 6, 2026). The letter characterizes these collectively as having required the enterprises and the Federal Home Loan Banks to comply with fair-lending and consumer-protection law, made that law enforceable through FHFA's authorities, set supervisory expectations, required regular reporting and proactive planning, and required officer certification of compliance.
It further notes that the CFPB withdrew a number of guidance documents relevant to PropTech and amended Regulation B to remove references to disparate impact at 91 FR 21620 (Apr. 22, 2026), a change NFHA opposed in a December 15, 2025 comment and is challenging in court (Source: nationalfairhousing.org). Parallel HUD actions cited are the withdrawal of fair housing and equal opportunity guidance at 91 FR 17291 (Apr. 6, 2026) and the proposed rule on the Fair Housing Act's disparate-impact standard at 91 FR 1475 (Jan. 14, 2026).
Provenance
Retrieved from NFHA's own domain, the canonical host for its comment letters, in full at twelve of twelve pages. The letter's date, addressee and subject match the account in Inside AI Policy, "Fair housing group argues against safe harbors from liability for any AI actor" (2026-08-21), which carried the item into New Developments Log/2026-08-21-0810-ai-developments.md from a lede-only retrieval.
Two details in that coverage are narrower or different in the letter itself. The submission is addressed to Ranking Member Maxine Waters, not to committee Democrats as a body — the RFI was issued by committee Democrats, and the response is directed to the Ranking Member. And the headline proposition, that no AI actor should receive a safe harbor, is stated in the letter as role-based allocation of liability across developers, deployers, vendors and users, which is not the same claim as uniform liability.
Relationships
- supports: AI Liability — argues against safe harbors and for role-differentiated allocation.
- supports: Algorithmic Accountability and Bias Audits — the PPM framework, LDA requirement and third-party audit proposals.
- contradicts: Executive Order 14365 — opposes the federal preemption policy the order directs.
- related: National Fair Housing Alliance, Federal Trade Commission (FTC), Financial Services — AI Deployment, AI Bias and Discrimination, AI and Tort Liability.