United States - AI Accountability Act (S. 2892)
Algorithmic Accountability Act (S. 2892, Senate version)
United States
RAI-US-NA-AAS2SXX-2023United States - AI Accountability Act (S. 2892) is Withdrawn in United States as of 9 Sep 2026, according to congress.gov.
BillGovernance and OversightAccountability and DocumentationTransparency and DisclosureS.2892 (Algorithmic Accountability Act of 2023) would require certain companies under FTC jurisdiction that develop or deploy automated decision systems for high‑impact "critical decisions" to conduct ongoing impact assessments, maintain documentation, submit summary reports to the Federal Trade Commission, and register selected information in a publicly accessible FTC repository. The bill directs the FTC to promulgate implementing regulations and creates staffing and resource authorizations to support oversight and enforcement.
Summary
The Algorithmic Accountability Act of 2023 (S.2892) is a Senate‑introduced bill (sponsored by Sen. Ron Wyden and cosponsored by a group of Senators) that would require the Federal Trade Commission (FTC) to promulgate regulations mandating impact assessments, documentation, reporting, and public repository disclosures for certain automated decision systems and "augmented critical decision processes." The bill defines covered entities by objective thresholds (revenue, equity value, identifying information on >1,000,000 consumers, or ownership relationships) and by whether the entity deploys an augmented critical decision process. "Critical decisions" include outcomes that have legal, material, or similarly significant effects on consumers in areas such as education, employment, essential utilities, family planning, financial services, healthcare, housing, and legal services; the FTC may expand or refine that list through rulemaking.
Under S.2892, the FTC must, within two years after enactment, issue regulations (in consultation with NIST, the National AI Initiative, OSTP, and other stakeholders) that require covered entities to conduct ongoing impact assessments of automated decision systems and augmented critical decision processes both prior to and after deployment. Impact assessments must include documentation of purpose and scope; data sources and quality; testing and evaluation; performance metrics; differential performance (disparate impacts) across protected and other characteristics; security and privacy protections; consumer rights mechanisms (contest/correct/appeal/opt‑out); and stakeholder engagement. Covered entities must maintain full impact assessment documentation for three years beyond the deployment period and must submit structured summary reports to the FTC — initial reports prior to deployment of new systems and annual summary reports thereafter — using machine‑readable formats prescribed by the FTC. The FTC must publish a limited, searchable public repository containing selected fields from summary reports (balancing public interest and commercial risk).
The bill establishes enforcement mechanisms by treating violations as unfair or deceptive acts/practices under the FTC Act and allowing FTC civil actions, administrative enforcement, and state attorney‑general suits. It authorizes creation of a Bureau of Technology at the FTC with additional staffing (minimum 50 personnel) and 25 additional enforcement personnel in the Bureau of Consumer Protection to carry out rulemaking, monitoring, outreach, and enforcement. The Commission must issue guidance and training materials to help entities determine covered status, and the bill requires periodic review of regulations at least every five years. The Act also requires coordination with other federal and state regulators and expressly does not preempt state or local law.
S.2892 aims to operationalize algorithmic accountability by making impact assessment, transparency, standardized reporting, stakeholder consultation, mitigation of material harms, and public visibility core compliance obligations for large or high‑impact entities that use AI and other automated decision systems for critical decisions. The bill balances disclosure with protections for trade secrets by limiting what is required to summary reports and by permitting certain non‑public retention of complete assessments, while enabling regulatory review by FTC, NIST, OSTP, and other relevant agencies to develop standards and future policy. If enacted, the statute would place the FTC at the center of federal oversight for high‑impact automated decision making, create new reporting burdens for covered entities, and authorize significant administrative and enforcement resources to oversee compliance.
Full article
Read full text ↗Overview
The Algorithmic Accountability Act of 2023 (S.2892) directs the Federal Trade Commission ("Commission" or "FTC") to promulgate rules requiring covered entities to perform and retain impact assessments for automated decision systems and "augmented critical decision processes"—systems that participate in making decisions with legal, material, or similarly significant effects on consumers. The bill (introduced Sept. 21, 2023) targets high‑impact uses such as education, employment, utilities, family planning, financial services, healthcare, housing, and legal services, and sets objective thresholds for "covered entity" status based on revenue, equity value, consumer data scale, or ownership relationships. The FTC is required to build a publicly accessible repository of summary report data to enable researchers, advocates, and consumers to study these systems while retaining mechanisms to protect sensitive information. For full statutory text and legislative status, see the official resources: Congress.gov - S.2892 text and the sponsor press release at Sen. Wyden press release. The bill also authorizes FTC staffing and coordination with federal standards bodies such as NIST and the Office of Science and Technology Policy (OSTP), reinforcing an evidence‑driven regulatory approach.
Definitions
S.2892 contains precise definitions to focus obligations: "automated decision system" (any computation‑based system used as basis for a decision, excluding passive infrastructure), "augmented critical decision process" (an automated decision system making a critical decision), "critical decision" (decisions affecting access, cost, terms, availability of services in specified categories), "covered entity" (entities under FTC jurisdiction that deploy augmented critical decision processes and meet financial or data scale thresholds), "impact assessment" (ongoing study of system impacts), "summary report" (machine‑readable subset of impact assessment data for submission), and "third‑party decision recipient" (any party beyond the consumer and covered entity that receives decision outputs). These definitions establish the statutory scope and the triggers for compliance obligations.
Governance and Institutional Framework
The Act centralizes regulatory authority at the Federal Trade Commission while mandating interagency consultation. Within two years of enactment the FTC must promulgate rules (per the Administrative Procedure Act) in consultation with the Director of NIST, the Director of the National AI Initiative (or equivalent office), OSTP, standards bodies, industry, academia, civil‑society advocates, and impacted communities. To operationalize oversight, S.2892 authorizes creation of a Bureau of Technology within the FTC and the appointment of at least 50 personnel dedicated to technical, standards, and research functions, plus 25 additional enforcement staff in the Bureau of Consumer Protection for civil enforcement. The FTC must also negotiate cooperation agreements with other Federal agencies and provide secure access to summary reports to NIST, OSTP, and agency heads for standards development. This structure pairs technical capacity building inside the FTC with cross‑agency alignment to support consistent rulemaking and enforcement. See the statutory text at Congress.gov - S.2892 text for governance provisions.
Key Focus Areas
S.2892’s substantive compliance regime concentrates on impact assessment, documentation, transparency, mitigation, and public reporting. Impact assessments must describe the decision purpose, datasets, data provenance and quality, training and labeling practices, technical and business performance metrics, benchmarking and test procedures, and evaluations of differential performance by race, sex, age, disability, religion, socioeconomic and other categories the FTC may specify. Assessments must also document stakeholder consultation, security and privacy protections, consumer rights pathways (contest, correct, appeal, opt‑out), and remediation plans for likely material negative impacts. Covered entities must keep full assessment documentation for three years beyond deployment, submit initial summary reports prior to a new system’s deployment, and submit annual summaries thereafter in a machine‑readable format defined by the FTC. The FTC will operate a searchable public repository containing a limited subset of summary report fields to promote research and public oversight while considering commercial risk. The bill recognizes tradeoffs between transparency and proprietary risk and explicitly allows the FTC to define repository content that balances those factors. The full bill text provides detailed lists of required elements for assessments and summary reports: Congress.gov - S.2892 text and the sponsor materials at Wyden press release provide context and supporting commentary.
Implementation Framework
The Commission has two years after enactment to adopt regulations under section 553 of title 5, U.S.C. The rulemaking must: define covered entity status and the calculation methods for consumer/data thresholds; set formats and machine‑readable standards for summary reports; prescribe impact assessment requirements and prioritization criteria for multiple systems; issue guidance and training to help entities determine covered status; and create the FTC public repository with search and accessibility features (consistent with the 21st Century Integrated Digital Experience Act). The FTC’s rulemaking must consider administrative burdens, the stage of system development, availability of data, existing privacy constraints, and feasibility. The Act authorizes appropriation of funds to hire specialized personnel, and allows the FTC to appoint certain staff without regard to civil service rules for rapid capability building. The FTC is tasked with producing guidance and training materials that will be periodically updated to reflect experience and stakeholder feedback. Key cross‑agency coordination is required with NIST, OSTP, and the National AI Initiative for standards alignment and technical guidance.
Monitoring and Evaluation
The Act requires ongoing monitoring through annual summary report submissions and retention of full impact assessment documentation for potential FTC inspection. The FTC must review regulations at least every five years and update them as appropriate. The public repository and machine‑readable reporting enable external researchers and civil society to study outcomes, enabling independent monitoring of compliance and system impacts. The bill also provides that the FTC make summary reports available in a private, secure manner to NIST and other Federal agencies to inform standards development. The combination of internal FTC oversight, periodic regulatory review, external researcher access, and interagency standardization is intended to create iterative evaluation loops to refine best practices and regulatory obligations over time.
Penalties, Liability, and Appeals
S.2892 treats violations of the Act and implementing regulations as violations of unfair or deceptive acts or practices under section 18(a)(1)(B) of the FTC Act, enabling the FTC to pursue civil enforcement actions including injunctive relief, civil monetary penalties (as authorized under the FTC Act and implementing statutes), disgorgement, and other remedies available under Federal law. The Act authorizes State attorneys general and other authorized state officials to bring civil actions consistent with the FTC’s powers. The Act also makes it unlawful to knowingly provide substantial assistance to a covered entity in violating the regulations. Civil venue and service provisions are specified, and the Act preserves entities’ rights to appeal administrative decisions under existing statutes. The bill does not create a private right of action in the federal statute but does not preclude state remedies; the FTC’s enforcement discretion and state AG litigation powers are the primary compliance incentives.
Relationship to Other Instruments
The statute is designed to complement existing laws and regulatory frameworks rather than preempt them. It requires coordination with existing privacy, consumer protection, civil rights, and sectoral regulations (for example, healthcare and financial rules), and explicitly states that it does not preempt State, tribal, city, or local laws. The Act references the FTC Act for enforcement mechanisms and directs secure sharing of summary reports with NIST and OSTP to inform standards and future regulation. It anticipates interplay with sectoral regulators where augmented critical decision processes overlap with domains already regulated by agencies such as HHS, CFPB, or state regulators, and requires cooperation agreements with relevant Federal agencies to clarify information‑sharing and enforcement responsibilities.
International Alignment
While focused on U.S. jurisdiction, S.2892 advances policy principles that align with international trends emphasizing risk‑based regulation, impact assessments, transparency, and technical standards (for example, the EU AI Act’s risk‑based approach and various OECD AI principles). The Act mandates consultation with technical standards bodies and NIST, which often coordinate with international standardization bodies, thereby encouraging harmonization of assessment methodologies, reporting formats, and testing practices. The public repository and machine‑readable reporting could facilitate cross‑jurisdictional research and comparability of enforcement approaches, though the bill leaves data export and international enforcement coordination to existing interagency mechanisms and bilateral/multilateral cooperation arrangements.
Implementation Timeline
| Milestone | Date/Period |
|---|---|
| Introduced in Senate | 2023-09-21 |
| Rulemaking deadline (promulgate regulations) | Within 2 years after enactment (statutory) |
| Effect of regulations | Effective 2 years after promulgation (statutory) |
| FTC Bureau staffing minimum appointment | Not later than 2 years after enactment (minimum 50 personnel) |
| Periodic regulatory review | At least every 5 years after promulgation |
Sources and References
| Source | Type |
|---|---|
| S.2892 - Algorithmic Accountability Act of 2023 (full text) | Primary Source |
| S.2892 - PDF (introduced version) | Primary Source |
| Sen. Wyden press release (sponsor materials) | Primary Source |
Requirements for a company
What an organisation has to do under United States - AI Accountability Act (S. 2892), at a glance. Not legal advice.
Related Regulations
United States - Algorithmic Accountability Act (H.R. 5628)
United States97% similar
United States - Preventing Algorithmic Collusion (S. 232)
United States90% similar
United States - Washington - Algorithmic Discrimination Bill (HB 1954)
United States90% similar
FTC AI Accuracy Policy Statement
United States89% similar
TRUMP AMERICA AI Act (Sen. Blackburn AI policy framework discussion draft)
United States89% similar
More AI regulation in United States
AI regulation in United States: full overview
- ABA Guidance on Lawyers' Ethical AI Use
- United States - AI Adoption in Government (S.1363/H.R.2575)
- United States - AI Risk Management Framework
- America's AI Action Plan
- AVM Quality Control Standards Rule
- United States - AI Bill of Rights
- United States - Chatbot Safety Guidelines
- Markey AI Chatbot Ad Inquiry
© Regulations.AI · updated on 20 Sep 2026 · reviewed against official sources on 9 Sep 2026