Relates to the use of automated lending decision-making tools by banks for the purposes of making lending decisions; allows loan applicants to consent to or opt out of such use.
This bill amends the New York Banking Law to regulate the use of automated lending decision-making tools by certain state-regulated banking entities and licensed lenders. It defines automated lending decision-making tools broadly to include software using algorithms, computational models, or artificial intelligence that materially automates or replaces human judgment in lending decisions, while excluding ordinary office software and internal administrative tools. The bill also defines lending decisions to include approvals, denials, offers, counteroffers, and changes to loan terms, including decisions affecting creditworthiness, loan amount, interest rate, collateral, repayment, and other material terms.
Covered entities that use these tools would be required to conduct annual impact assessments, and also conduct an assessment before making any material change to a tool that could affect its output. Those assessments must evaluate the tool’s objectives, design, training data, accuracy, fairness, bias, discrimination, cybersecurity, privacy, public health and safety risks, misuse risks, and data practices. A summary report must be posted publicly on the entity’s website before the tool is used and updated after each subsequent assessment, while the full assessment must be retained for seven years and provided to the Department of Financial Services upon request.
The bill would add a new section 103-a to the Banking Law and create a new compliance framework for state-supervised banks, foreign banking corporations licensed in New York, interstate branches authorized in New York, and licensed lenders. It would require notice to loan applicants when automated tools are used, including advance notice of the tool’s use and criteria, disclosure of data sources and retention policies, and post-denial notice of the reason for denial where practicable. Applicants denied based on incorrect personal information would have 30 days to correct the information and appeal.
The bill also gives the superintendent of financial services enforcement and oversight authority, including the power to investigate, compel records, and require additional reporting if discriminatory or biased outcomes are found. If an impact assessment identifies discriminatory or biased results, the covered entity must report that finding within 30 days, and the department must direct the entity to stop using the tool and any information produced by it. The bill would take effect 90 days after becoming law and would apply only to the covered entities defined in the statute, not to national banks and certain other federally chartered institutions.
The available legislative history suggests generally favorable committee sentiment toward the bill, as it was reported favorably from the Senate Banks Committee and later advanced after amendment. The recorded committee vote was 4-2 in favor, indicating support but not unanimity. The bill’s caption also reflects a consumer-protection framing, with an emphasis on consent and opt-out rights for applicants, which likely contributed to its positive reception among supporters.
The main points of contention appear to center on the scope of regulation over artificial intelligence and automated underwriting, especially the burden of annual impact assessments, public reporting, and applicant notice requirements. Supporters are likely focused on preventing discrimination, increasing transparency, and giving borrowers a meaningful way to challenge inaccurate or biased automated decisions. Opponents or skeptics may be concerned about compliance costs, operational burdens, disclosure of proprietary model information, and the possibility that the bill could slow or discourage use of lending technology. The bill also expressly excludes federally chartered institutions, which may reflect both jurisdictional limits and a potential area of policy debate about uneven coverage.