highbeamSubscribe

Policy brief 27 · Economic Competitiveness

Disclosure of Organizational AI Use

When AI shapes a decision about you, you deserve to know.

All policy briefs

A lender that turns down a loan must give specific reasons, even when a scoring model made the call.1 An employer whose software screens out a résumé owes no such notice under federal law, and people cannot challenge what they cannot see. Congress should require plain notice when AI stands in for a person or shapes a consequential decision, a human who can fix errors, and a public register.

The Problem

Seven in ten Americans say it is extremely or very important that their doctor tell them when AI is used in their care, and 53% say they have little or no say over whether it is.2 The same question now reaches hiring, where companies increasingly use automated tools to scan résumés and analyze video interviews,3 and it reaches lending, insurance, and public benefits.

Much of this can help. But a person who doesn't know software shaped a decision can't ask what it saw, correct a wrong input, or reach someone with the power to change the result. Federal credit law already answers this for loans: a lender must state specific reasons for a denial, and saying the applicant "failed to achieve a qualifying score" is not enough.1 Outside credit, federal law requires no such notice. States are filling the gap unevenly, and three failures stand out:

  1. Notice is patchy. Illinois began requiring employers to notify workers of AI use in hiring, promotion, discipline, and other employment decisions only in January 2026;4 Colorado's broader notice law takes effect in January 2027.5
  2. No one can fix it. Federal agencies using high-impact AI must, where appropriate, give affected people "a timely human review and a chance to appeal."6 Private employers, lenders, and insurers face no equivalent federal rule.
  3. Rules go unenforced. New York City has required notice of AI hiring tools since 2023, yet its enforcer received two complaints in two years. Where the city found one problem among 32 companies, state auditors found at least 17 potential violations.3

Why legislation: Washington already holds itself to this standard. A 2022 law requires federal agencies to publish inventories of their AI uses,7 and a bipartisan bill requiring agency notices and appeals was reported by a Senate committee in 2023.8 The FTC can act against a company that lies about its use of AI;9 no federal law requires a company to disclose it. Institutions that hand judgment to a machine remain answerable for it.

The Solution

A four-step staircase: each step stands alone, and each step up adds accountability. Scope: decisions about employment, credit, housing, insurance, education admissions, health services, and public benefits, plus customer service where AI stands in for a person; spellcheck, calculators, and other incidental tools are exempt, and federal agencies are covered by a parallel title.

Step 1 — Say when it's a machine. When AI stands in for a person in a service interaction, say so at the start. Utah has required this since 2024 in licensed professions, and on request in other consumer dealings.10

Step 2 — Give notice at the point of decision. Before AI materially influences a covered decision, state its role, the responsible organization, the kinds of information used, and how to seek review. Repeat the notice with any adverse decision, saying whether a person decided with AI's help or the process was substantially automated. Colorado will require a plain-language account within 30 days of an adverse outcome.5 A person can't appeal a decision they don't know a machine made.

Step 3 — Put a human within reach. Give each affected person a contact who can investigate disputed inputs, explain the process, and change the outcome, and require vendors to give deployers the documentation this takes. Colorado's new law grants similar rights to correct data and request human review.5

Step 4 — Publish the register. Organizations with 500 or more employees publish an annual register of their consequential AI uses and the business function accountable for each, updated when new uses begin. Individual notices remain required at any size. Federal agencies already publish inventories like this.7

Where to start: Step 1 is the floor; it costs a sentence. Step 2 is the heart of the proposal.

Administration and enforcement: The FTC writes common notice rules within 12 months, with compliance at 18. Sector regulators enforce equivalent rules in their fields, and OMB administers the agency title. Civil penalties and corrective orders cover material failures and false notices, with a one-time cure for minor formatting errors.

Risks and Mitigations

  • Compelled speech: A notice that software screened an application is "purely factual and uncontroversial information" about a service, the kind the Supreme Court has let states require of commercial speakers.11 A public register reaches further, and the risk remains.
  • Notice fatigue: Labels on everything would blur the ones that matter. Americans draw the line themselves: 81% want to be told if AI helps diagnose them, 56% if it schedules their appointment.2 Covering only consequential uses keeps notices meaningful, though the borders will be argued.
  • Trade secrets and rubber stamps: Organizations disclose the use and who answers for it, not model weights or fraud rules; regulators can inspect confidential files. Reviewers must see the records and hold power to change the result. Even real review can err, so anti-discrimination law stays essential.

Similar Bills

Fit measures similarity to this proposal's mechanisms: High = direct precedent; Partial = useful component with material differences; Related = adjacent approach.

Federal

Proposal or bill Relevant provisions and fit Fit
S. 1865 — Transparent Automated Governance Act
Peters (D-MI), Braun (R-IN), Lankford (R-OK)
118th Congress · Reported Aug. 22, 2023; not enacted		 Reported §3 directs OMB guidance requiring agencies to give plain-language notice at the point of interaction and when a critical decision issues, plus appeals and alternative human review. Direct government analogue for Steps 1–3; covers agencies only. High
H.R. 6371 — No Robot Bosses Act
Bonamici (D-OR), Deluzio (D-PA), Moylan (R-GU)
Referred to committee · Dec. 3, 2025		 §3 requires employers to disclose automated decision systems, corroborate their outputs with human oversight, document each decision within 7 days, and allow disputes and appeals to a human. Closest employment precedent for Steps 2 and 3; also bars sole reliance on the system and requires testing, beyond this draft. Senate version: S. 4833. High
S. 2164 — Algorithmic Accountability Act of 2025
Wyden (D-OR) + 7 Democratic cosponsors
Referred to committee · June 25, 2025		 Requires impact assessments of automated systems used in critical decisions, summary reports to the FTC, and a public repository (§6). Precedent for Step 4's register, with much heavier assessment duties. House version: H.R. 5511; earlier version: H.R. 5628 (118th). Partial
H.R. 3831 — AI Disclosure Act of 2023
Torres (D-NY)
118th Congress · Introduced June 5, 2023; not enacted		 Would require a disclaimer on AI-generated output. Supports identifying AI involvement, as in Step 1; output labeling is narrower than explaining a decision. Related

State

Proposal or bill Relevant provisions and fit Fit
Colorado — SB 26-189 / Chapter 131
Signed May 14, 2026 · Effective Jan. 1, 2027
Attorney general rulemaking pending		 Repeals and reenacts the 2024 Colorado AI Act. Requires notice before automated decision-making technology materially influences a consequential decision, a plain-language explanation within 30 days of an adverse outcome, correction of inaccurate data, and human review on request. Closest precedent for Steps 2 and 3; no public register. High
Illinois — HB 3773 / Public Act 103-0804
Signed Aug. 9, 2024 · Effective Jan. 1, 2026		 Amends the Illinois Human Rights Act to bar discriminatory employment uses of AI and to require notice to employees when AI is used in covered decisions. Direct workplace precedent for Step 2; no human-review right or register. High
New York City — Local Law 144 of 2021
Enacted Dec. 11, 2021 · Enforced from July 5, 2023		 Requires annual bias audits of automated hiring tools, public audit summaries, and notice to candidates at least 10 business days before use, with a chance to request an alternative process. Precedent for Steps 2 and 4; a December 2025 state audit found enforcement ineffective. High
Utah — SB 149 (2024)
Signed March 13, 2024 · Effective May 1, 2024; original act compared		 Requires prominent disclosure of generative AI in regulated occupations and, when asked, in other consumer interactions. Direct precedent for Step 1; narrower than decision notices. High
Connecticut — SB 1103 / Public Act 23-16
Approved June 7, 2023		 Requires state agencies to inventory their AI systems and publish the inventories, with impact assessments. Public-register precedent for Step 4; covers government use only. Partial

What this adds: Colorado, Illinois, and New York City each cover a piece: notice, employment, or hiring audits. This proposal sets one national floor across consequential decisions, adds a human with real authority to fix errors, and makes compliance visible through a public register. A separate proposal restricts chatbots that pose as people; this one covers organizations that use AI on the people they serve.

Notes

  1. Regulation B, 12 C.F.R. § 1002.9(a)(2), (b)(2), implementing the Equal Credit Opportunity Act. The CFPB withdrew its guidance applying this rule to complex algorithms (Circular 2022-03) on May 12, 2025; the regulation is unchanged. ↩ ↩2

  2. Pew Research Center, "Americans Want Transparency When AI Is Used in Their Healthcare," August 25, 2026. Survey of 3,488 U.S. adults, June 22–28, 2026. ↩ ↩2

  3. Office of the New York State Comptroller, Enforcement of Local Law 144 – Automated Employment Decision Tools, Report 2024-N-6, December 2025, pp. 1–2. Audit period July 2023–June 2025; the 17 are potential violations. ↩ ↩2

  4. Illinois Public Act 103-0804 (HB 3773), amending 775 ILCS 5/2-102(L), effective January 1, 2026. ↩

  5. Colorado SB 26-189, Chapter 131, Session Laws 2026, §§ 6-1-1704 to 6-1-1705, signed May 14, 2026, effective January 1, 2027. Implementing rules pending as of September 2026. ↩ ↩2 ↩3

  6. Office of Management and Budget, Memorandum M-25-21, April 3, 2025, p. 17 (minimum practices for high-impact AI). ↩

  7. Advancing American AI Act, Pub. L. No. 117-263, div. G, § 7225 (2022); OMB M-25-21, p. 12, requires public inventories at least annually. ↩ ↩2

  8. S. 1865, Transparent Automated Governance Act, 118th Cong. § 3 (reported text, August 22, 2023). Sponsored by Sens. Peters (D-MI), Braun (R-IN), and Lankford (R-OK). ↩

  9. Federal Trade Commission Act, 15 U.S.C. § 45(a)(1), barring "unfair or deceptive acts or practices." ↩

  10. Utah SB 149 (2024), enrolled text § 13-2-12(3)–(5), effective May 1, 2024. Original act; later amendments not reviewed. ↩

  11. Zauderer v. Office of Disciplinary Counsel, 471 U.S. 626 (1985), upholding a required disclosure of factual information about the terms of a lawyer's services. ↩