Algorithmic Accountability Act of 2025
- Sponsor
- Rep. Clarke, Yvette D. [D-NY-9]
- Committees
- Energy and Commerce Committee (primary)
- Last action
- Sep 19, 2025
Bottom line
This bill establishes a comprehensive federal framework for algorithmic accountability, mandating impact assessments and public reporting for significant algorithms used in critical decision-making by large entities.
What it actually does
This bill directs the Federal Trade Commission (FTC) to establish regulations requiring large entities that deploy certain "covered algorithms" to conduct impact assessments. These assessments must evaluate privacy risks, performance, fairness, and consumer rights, and identify and mitigate negative impacts. Covered entities must submit summary reports to the FTC, and the FTC will create a public repository of limited information from these reports.
Proponents argue
Supporters argue the bill is crucial for protecting consumers from potential harms, biases, and discriminatory outcomes of increasingly prevalent artificial intelligence and algorithmic systems. They contend it promotes transparency, fairness, and accountability in algorithmic deployment, ensuring that powerful technologies are developed and used responsibly, especially in critical areas like employment, housing, and healthcare.
Opponents contend
Potential opponents might argue that the bill imposes significant regulatory burdens and compliance costs on businesses, potentially stifling innovation and competitiveness in the rapidly evolving AI sector. They might also raise concerns about the disclosure of proprietary information through summary reports, even if limited, and the potential for overreach by the FTC into complex technological development.
The bill is moderately complex due to its technical definitions and detailed regulatory requirements, necessitating a careful read to fully grasp its implications.
Section 2(5), 2(6), 2(7)
Definitions of Covered Algorithms, Entities, and Critical Decisions
This provision defines the scope of the bill by specifying which algorithms are subject to regulation, which entities must comply, and what types of decisions are considered 'critical.' A 'covered algorithm' includes AI, machine learning, and similar processes affecting 'consequential actions.' A 'covered entity' is generally a large company (over $50M gross receipts or $250M equity, or using identifying info for over 1M consumers). 'Critical decisions' are those with significant effects on a consumer's life, such as in employment, housing, healthcare, and financial services.
Supporters argue
Proponents argue these definitions are essential to clearly delineate the scope of the Act, ensuring that significant algorithmic systems and the entities deploying them are held accountable, particularly when their decisions have substantial impacts on individuals.
Critics contend
Critics might argue that the thresholds for 'covered entities' are too broad or too narrow, potentially capturing entities that should not be regulated or missing others that should. The definitions of 'covered algorithm' and 'critical decision' could also be seen as overly vague, leading to uncertainty and compliance challenges.
Tradeoffs
The tension lies in balancing comprehensive coverage to address potential harms with providing clear, manageable boundaries for regulated entities. Broad definitions ensure wide applicability but may create ambiguity, while narrow definitions offer clarity but risk overlooking emerging issues.
Section 3(b)
FTC Mandate for Algorithmic Impact Assessment Regulations
This section directs the Federal Trade Commission (FTC) to promulgate regulations within two years. These regulations will require covered entities to perform impact assessments of their covered algorithms both before and after deployment. Entities must maintain documentation of these assessments, disclose their 'covered entity' status, and submit annual and initial summary reports to the FTC. The FTC is also tasked with establishing definitions for key terms and guidelines for compliance.
Section 4(a)
Detailed Requirements for Algorithmic Impact Assessments
This section outlines the comprehensive requirements for the impact assessments that covered entities must perform. These include evaluating existing decision-making processes, consulting with internal and external stakeholders, conducting ongoing testing for privacy risks and performance (including differential performance across demographic groups), training employees, assessing guardrails for algorithm use, documenting data sources and quality, evaluating consumer rights (notice, opt-out, explainability, contestability), identifying and mitigating negative impacts, and documenting the development process. Entities must also document any requirements that were infeasible to comply with, along with the rationale.
Section 5, Section 6(b)
Summary Reports to FTC and Publicly Accessible Repository
Covered entities are required to submit summary reports to the FTC, detailing key information from their impact assessments, including entity contact, critical decision description, intended purpose, stakeholder consultations, testing results (including differential performance), guardrails, data documentation, transparency measures, identified negative impacts, and reasons for infeasible requirements. The FTC must then develop and maintain a publicly accessible repository containing a limited subset of this information, such as the entity's identity, the critical decision, publicly stated prohibitions, data sources, performance metrics, and links for consumer recourse.
Section 8(a), 8(b)
Establishment of FTC Bureau of Technology and Additional Personnel
This section establishes a new Bureau of Technology within the Federal Trade Commission, to be headed by a Chief Technologist. This Bureau will advise the Commission on technological aspects of its functions, aid in enforcement of the Act, and provide technical assistance. The bill authorizes the appointment of at least 50 personnel with expertise in fields like technology, digital design, civil rights, software engineering, and machine learning for this Bureau. Additionally, it authorizes 25 more personnel for the Division of Enforcement within the Bureau of Consumer Protection to support enforcement efforts.
Section 9
Enforcement by FTC and States
This section establishes the enforcement mechanisms for the Act. Violations of the Act or its regulations will be treated as unfair or deceptive acts or practices under the Federal Trade Commission Act, granting the FTC its full range of enforcement powers, including penalties and consumer redress. Additionally, state attorneys general are empowered to bring civil actions on behalf of their residents if they believe state interests are threatened by a violation. States must notify the FTC before initiating such actions, and the FTC retains the right to intervene.
Section 11
No Preemption of State and Local Laws
This section explicitly states that nothing in the Act should be construed to preempt any existing or future State, tribal, city, or local law, regulation, or ordinance. This means that state and local governments retain the authority to enact their own laws regarding algorithmic accountability, even if they are more stringent than the federal requirements established by this Act.
This subsection mandates that the dollar amounts specified for determining 'covered entity' status (e.g., $50,000,000 in gross receipts or $250,000,000 in equity value) shall be increased by the percentage increase in the consumer price index for all urban consumers for any fiscal year after the first fiscal year following enactment.
Section 2(6)(B)
Why it matters:This is standard legislative practice for ensuring that monetary thresholds in statutes remain relevant over time by adjusting for inflation, rather than being a deliberate attempt to obscure the provision.
Case for: Proponents would argue that this inflation adjustment is a necessary and responsible measure to ensure the bill's applicability remains appropriate over decades, preventing the thresholds from becoming outdated and inadvertently capturing smaller entities over time due to inflation.
Case against: No specific opposition is typically raised against this provision because it is a standard technical adjustment. However, some might argue that the specific inflation index used or the frequency of adjustment could be debated.
Estimated impact: Ensures that the financial thresholds for 'covered entity' status will increase over time, maintaining the bill's focus on larger entities as the economy grows.
This subsection explicitly states that nothing in the Act requires a covered entity to share with or disclose to the Commission or the public any information contained in an impact assessment, except for the information required in the summary report under Section 3(b)(1)(D) or (E).
Section 4(c)
Why it matters:This provision is likely placed here to provide a clear boundary for disclosure requirements, addressing potential industry concerns about proprietary information while still mandating public-facing summary reports. It's a key compromise.
Case for: Proponents argue this clause is crucial for protecting covered entities' proprietary information, trade secrets, and competitive advantages, which might be contained in the full impact assessment. It balances transparency with the need to safeguard sensitive business data, encouraging compliance without undue commercial risk.
Case against: Critics might argue that this limitation reduces the overall transparency and accountability of the process, as the public and even the FTC (beyond the summary) will not have access to the full details of the assessments. This could hinder deeper scrutiny and independent verification of claims made in summary reports.
Estimated impact: Limits the scope of public and regulatory disclosure to summary information, protecting detailed internal assessments.
The bill requires covered entities, to the extent possible, to meaningfully consult with relevant internal and independent external stakeholders (including representatives of impacted groups, civil society, and technology experts) as frequently as necessary during impact assessments. Section 4(a)(2) further details documentation requirements for these consultations, including points of contact, dates, terms, materials interacted with, and recommendations made (and rationale for non-use).
Section 3(b)(1)(G) and Section 4(a)(2)
Why it matters:This level of detail is typical for legislative instructions to a regulatory agency, ensuring that the agency's rulemaking will incorporate specific procedural safeguards. It's a critical operational detail rather than a hidden policy.
Case for: Proponents argue that 'meaningful consultation' with diverse stakeholders, especially impacted communities, is vital for identifying unforeseen biases and harms that technical teams might miss. Documenting this process ensures accountability and transparency in the assessment itself, leading to more robust and equitable algorithms.
Case against: Some covered entities might view the detailed consultation and documentation requirements as an additional administrative burden, potentially slowing down development cycles. They might argue that internal ethics reviews are sufficient or that external consultations can be difficult to manage and integrate effectively.
Estimated impact: Mandates a participatory approach to algorithmic impact assessment, integrating diverse perspectives into the development and deployment process.