The House and Senate are advancing legislation that would let financial firms sidestep federal consumer protections, civil rights laws, and financial stability rules by deploying artificial intelligence. The bills create a regulatory sandbox where companies that use AI can apply for broad exemptions from existing law - and regulators are required to approve those exemptions if they are "more likely than not" to meet loose guidelines.
The stated goal is to let financial firms experiment with AI without "unnecessary or unduly burdensome regulations." But in practice, the legislation encourages companies to adopt risky AI systems specifically to secure waivers from oversight. Practices that would be illegal if carried out by a person would become permissible if performed by an AI system.
"It would let financial firms off the hook for wrongdoing merely by the virtue of using AI and let the firms capture all the benefits of AI and force the public to bear all of the risks," the policy brief states.
How the sandbox works
Regulated financial firms can apply for waivers from federal laws and regulations for "AI test projects." The legislation does not limit the scope of those waivers. A company could request total exemption from consumer protection, fair lending, risk management, capital requirements, investor protection, or market integrity rules.
Regulators are required to approve applications if they are "more likely than not" to meet the criteria. That standard is exceptionally low. It means agencies must approve a project even if it carries real risk of failing those weak guidelines.
Once approved, regulators cannot enforce federal law against the project - only the company's own proposed "alternative compliance strategy" applies. The legislation includes no requirement that companies disclose the use of AI to customers, reveal what personal data is collected, or offer a way to opt out.
No protections for racial bias or consumer harm
The legislation makes no reference to core civil rights obligations, including fair lending and community reinvestment requirements. This matters because AI-enabled underwriting and credit decision systems are known to amplify racial disparities in loan approval, pricing, and terms.
"Many studies have documented that AI enabled underwriting and decision-making systems amplify existing racial disparities in the approval, pricing, and terms of loans," the analysis states.
Consumer protections are equally thin. The legislation offers no remedy for people harmed by AI errors or model failures. It does not mandate ongoing testing or auditing to check whether test projects develop emerging risks. The injunctive relief provision lets regulators sue only when an AI project poses "immediate danger to consumers or investors" - a standard that appears nowhere else in federal financial law and would be difficult to invoke before damage is done.
The House bill lets agencies issue cease and desist orders only when a project is "causing unmitigable or irreparable harm." At that point, the harm is already complete.
Risks to financial stability
The legislation also weakens oversight of AI used in trading, risk management, and asset allocation. Banks relying on opaque black-box AI models may underestimate risk - similar to how subprime mortgage models appeared safe before the 2008 crisis. The widespread use of a small number of AI models across the industry could amplify correlated trading strategies and inflate asset bubbles.
Companies must attest that their AI test project "would not pose systemic financial risk," but the legislation offers no standard for evaluating that claim. Regulators can only stop a project that is already causing "unmitigable or irreparable harm to financial stability."
For finance professionals, the implications are direct. The legislation creates a pathway for competitors to operate under different rules - potentially making loans, pricing securities, or managing risk without the safeguards that govern traditional operations. It also concentrates model risk: if multiple institutions rely on the same few AI systems, a failure in one could cascade across the system.
Your membership also unlocks:
