Algorithmic Accountability Act: What It Requires

Last verified: September 20, 2026. The Algorithmic Accountability Act is a federal bill — it has never been enacted. That distinguishes it from every state AI law covered elsewhere on this site: Colorado’s SB 26-189, Utah’s AI Policy Act, Texas’s TRAIGA, and California’s SB 53 are all current law today. The Algorithmic Accountability Act is not, and has never been, despite being introduced in four separate Congresses since 2019. Its most recent version — the Algorithmic Accountability Act of 2025 — is pending in committee in both chambers as of this writing, with no floor vote scheduled in either.

A Bill Introduced Four Times, Never Enacted

The bill is Senator Ron Wyden’s (D-OR), usually co-led with Senator Cory Booker (D-NJ) in the Senate and Representative Yvette Clarke (D-NY) in the House. Every version has died in committee:

Congress / Year	Bill(s)	Sponsor(s)	Introduced	Outcome
116th (2019)	S. 1108 / H.R. 2231	Sen. Wyden; Rep. Clarke	April 10, 2019	Referred to committee; no vote
117th (2022)	S. 3572	Sen. Wyden	February 3, 2022	Died in committee, no vote
118th (2023)	Reintroduced (Algorithmic Accountability Act of 2023)	Sens. Wyden, Booker; Rep. Clarke	September 22, 2023	Referred to committee; no vote
119th (2025) — current	S. 2164 / H.R. 5511	Sen. Wyden (S. 2164); Rep. Clarke (H.R. 5511)	June 25, 2025 (Senate) / September 19, 2025 (House)	Pending in committee

As Senator Wyden put it when reintroducing the bill in 2023, neither the 2019 nor the 2022 version “progressed past the committee level.” The pattern has continued: the 2025 version — the one this guide covers — has not advanced either. Independent bill-tracking analysis from GovTrack puts S. 2164’s odds at roughly 3% of passing committee and 1% of eventual enactment; H.R. 5511 is tracked at roughly 2% and 0% respectively. Treat “algorithmic accountability act,” “algorithmic accountability act of 2025,” and bare “algorithmic accountability” as the same current, unenacted proposal — they all refer to this bill in its present form, not to a law already on the books.

What the Bill Would Require, If Passed

The Senate and House versions are close but not textually identical — a normal state of affairs for companion bills still in committee. S. 2164’s official title directs the FTC “to require impact assessments of automated decision systems and augmented critical decision processes,” using the term automated decision system throughout. H.R. 5511 instead defines a covered algorithm — “a computational process derived from machine learning, natural language processing, artificial intelligence techniques, or other computational processing techniques of similar or greater complexity” that makes or materially assists a decision. The two chambers have not harmonized on a single core term, which is itself a sign of how early-stage the bill remains.

Both versions center on a critical decision (S. 2164) or consequential action (H.R. 5511) — a decision with legal or similarly significant effects on a consumer’s access to, cost of, or terms for education, employment, housing, credit or other financial services, healthcare, insurance, legal services, or utilities. Coverage does not apply to every company; H.R. 5511 sets it by entity size: roughly $50 million in average annual revenue or $250 million in equity value for a deploying entity, a lower $5 million/$25 million threshold for entities that develop covered algorithms for larger deployers, and a separate trigger for any entity holding identifying data on 1 million or more consumers or devices used to develop a covered algorithm.

For a covered entity, the bill’s core mechanism — the one this guide is built around — is the impact assessment: an ongoing study and evaluation of the algorithm’s effect on consumers, required before deployment and re-evaluated afterward. Per the bill text, the assessment has to document, among other things:

The process or system the algorithm replaces, and known harms of the prior approach
Consultation with affected stakeholders
Privacy risks and privacy-enhancing measures
How the algorithm performs in testing versus in actual deployment
Differential performance across demographic groups
Data sourcing and data-quality documentation
Consumer-facing rights: notice, opt-out, transparency, and appeal mechanisms
Identified negative impacts and the steps taken to mitigate them

A summary version of the assessment must be filed with the FTC before deployment, with an annual update afterward. The bill requires the FTC to stand up a public, searchable repository of these summaries — updated quarterly, and including the entity’s identity, the specific decision the algorithm affects, its data sources, and its performance results — and to build out enforcement capacity for the job: a new Bureau of Technology headed by a Chief Technologist, required to reach at least 50 staff within two years, plus 25 more in the Bureau of Consumer Protection’s enforcement division. That is where the commonly cited figure of roughly 75 new FTC staff comes from.

Where It Stands Right Now

S. 2164 was referred to the Senate Committee on Commerce, Science, and Transportation on introduction (June 25, 2025) and has seven Democratic cosponsors. H.R. 5511 was referred to the House Committee on Energy and Commerce (September 19, 2025) and has 32 cosponsors, also entirely Democratic. Neither bill has had a committee markup or floor vote scheduled as of this writing, and neither has a Republican cosponsor in a Congress where the majority controls the calendar in both chambers — the structural reason GovTrack’s models put its odds so low. None of that makes the bill irrelevant to compliance planning: its requirements, especially the impact-assessment content list above, closely track what several state laws and the EU AI Act already require in substance, even though this specific federal version remains proposed.

How It Compares to What’s Already Enacted

Every other AI-related law on this site is current, enforceable law. The Algorithmic Accountability Act is not — it is the proposal against which those enacted laws can be measured:

Law	Level	Status	Core mechanism
Algorithmic Accountability Act of 2025 (S. 2164 / H.R. 5511)	Federal	Proposed — pending in committee	Pre- and post-deployment impact assessments, summary reports to a public FTC repository
Colorado SB 26-189	State	Enacted — effective January 1, 2027	Developer/deployer duties for “automated decision-making technology” in consequential decisions
Utah AI Policy Act	State	Enacted	Disclosure-focused; narrower than Colorado’s law
Texas TRAIGA (HB 149)	State	Enacted — effective January 1, 2026	Bans specific AI uses by intent, rather than regulating by decision type or model scale
California SB 942	State	Enacted	AI content-provenance and transparency disclosures (a different target than impact assessments)
California SB 53	State	Enacted	Frontier-model transparency reports and incident disclosure — scoped by training compute, not decision type
EU AI Act, Article 27	Supranational	Enacted	Fundamental Rights Impact Assessment, required of deployers of high-risk AI systems before first use

The closest existing analogue to what the Algorithmic Accountability Act would create is the EU AI Act’s Article 27: it also requires an impact assessment before deployment, covering affected groups, identified risks, and mitigation measures, filed with a regulator (a market surveillance authority rather than the FTC). The state laws in this comparison take a different shape. Colorado’s SB 26-189 and Utah’s law regulate “automated decision-making technology” the way the federal bill would, but through documentation and notice duties rather than a mandatory, regulator-facing impact assessment. Texas’s TRAIGA does not use an impact-assessment or high-risk-decision framework at all — it bans specific uses outright. California’s SB 53 is scoped by training compute and targets frontier model developers specifically, a different regulatory question than “was this hiring or lending algorithm tested for bias,” which is what the Algorithmic Accountability Act and Colorado’s law are both actually about.

The Federal Layer This Cluster Was Missing

Until recently, every piece of federal AI policy covered here came from the executive branch rather than Congress — see our guide to Trump’s AI executive orders, EO 14179 and EO 14365. That is a fundamentally different mechanism: an executive order takes effect the moment it is signed and can be revoked by the next administration with a stroke of a pen (as EO 14179 itself did to its Biden-era predecessor), while a bill like the Algorithmic Accountability Act has to clear committee, pass both chambers, and be signed into law before it has any legal force at all — and, as the legislative history above shows, this particular bill has not gotten past the first of those steps in four attempts across six years. Reading the two guides together gives the fuller federal picture this cluster previously lacked entirely: what the executive branch has already done unilaterally, and what Congress has repeatedly tried, and so far failed, to enact.

What This Means for NIKOLAI

The impact assessment is the Algorithmic Accountability Act’s core mechanism: a defined, documented procedure that produces evidence about a system’s behavior before it is relied on for a consequential decision. CASRAI’s own NIKOLAI dictionary is scoped differently — it maps terminology for frontier-model catastrophic-risk governance, not consumer-protection algorithmic-bias review, so it has no element built specifically for a civil-rights-style impact assessment. The closest structural analogue in NIKOLAI’s vocabulary is Evaluation, in Track N5, Evidence and Evaluations: NIKOLAI’s proposed reading is “a defined procedure that produces evidence about a model’s capabilities, propensities, or safeguard effectiveness, recorded with its purpose, method type, and reporting format as a distinct instance.” That is the same underlying shape as what the Algorithmic Accountability Act would require — a structured, repeatable, documented assessment, reported in a consistent format to an outside party — even though the two come from different regulatory traditions and NIKOLAI’s own tracks were not built with algorithmic-bias assessments in mind. As with every other NIKOLAI crosswalk, this reading is CASRAI’s own interpretation, not a mapping any lab, regulator, or bill sponsor has confirmed: it is a shadow mapping unless and until an organization files its own Mapping Declaration for it. Teams that end up building internal tooling to track and produce algorithmic impact assessments under a law like this one can also pull NIKOLAI’s element and track definitions directly through CASRAI’s public v1 REST API (documented at /implement/api-reference), rather than re-typing them by hand into their own documentation systems.

Frequently Asked Questions

Is the Algorithmic Accountability Act a law right now?

No. It is a bill — S. 2164 in the Senate and H.R. 5511 in the House, both introduced in 2025 and both still pending in committee. It has been introduced four times since 2019 and has never passed either chamber.

What is the difference between “Algorithmic Accountability Act,” “Algorithmic Accountability Act of 2025,” and “algorithmic accountability” as search terms?

They all refer to the same current proposal. “Algorithmic Accountability Act of 2025” is the formal short title of the bill as reintroduced in the 119th Congress; “Algorithmic Accountability Act” and “algorithmic accountability” are the ways people commonly search for or refer to the same bill, not a separate law.

What would the bill actually require companies to do?

Covered entities — roughly, larger companies that deploy or develop qualifying automated systems used in decisions about education, employment, housing, credit, healthcare, insurance, legal services, or utilities — would have to conduct and document an impact assessment before deployment and update it annually, then file a summary with the FTC, which would maintain a public, searchable repository of those summaries.

How is this different from Colorado’s or Utah’s AI laws, if they cover similar ground?

The state laws are enacted and enforceable now; the federal bill is not. Structurally, Colorado’s SB 26-189 and Utah’s AI Policy Act impose disclosure and documentation duties on developers and deployers of automated decision-making technology, while the federal bill would go further by requiring a formal impact assessment filed with a federal regulator and made public through an FTC repository.

Does the bill apply to every company that uses AI?

No. The House version sets revenue and equity thresholds — roughly $50 million in average annual revenue or $250 million in equity value for entities that deploy covered algorithms, and lower thresholds for entities that develop them for larger deployers — plus a separate trigger for entities holding identifying data on 1 million or more consumers or devices.

What are the odds this bill becomes law?

Low, on the evidence so far. GovTrack’s independent prediction model puts S. 2164 at roughly a 3% chance of passing committee and a 1% chance of enactment, and H.R. 5511 at roughly 2% and 0%. Both bills have exclusively Democratic cosponsors in a Congress where that alone has historically been a strong predictor of a bill not advancing.

The Algorithmic Accountability Act: What the Federal Bill Would Require