US Algorithmic Accountability Act (Proposed)

Summary

The Algorithmic Accountability Act represents the most comprehensive attempt yet to regulate automated decision-making systems in the United States. Introduced in 2023, this proposed legislation would require companies using high-risk AI systems to conduct detailed impact assessments, similar to environmental impact studies. Unlike voluntary frameworks or industry self-regulation, this bill would create legally binding obligations for algorithmic auditing, with enforcement powers vested in the Federal Trade Commission. If passed, it would fundamentally shift how companies deploy AI in critical areas like hiring, lending, healthcare, and criminal justice.

The backstory: Why now?

This isn't the first attempt at algorithmic accountability legislation in the US. Earlier versions were introduced in 2019 and 2022 but failed to gain traction. The 2023 version emerges in a dramatically different landscape - after high-profile AI failures in hiring algorithms, concerns about facial recognition bias, and the explosive growth of generative AI. The bill's sponsors learned from EU GDPR implementation and the EU AI Act negotiations, crafting requirements that are more specific and actionable than previous attempts.

The timing coincides with growing bipartisan concern about AI's impact on American workers and consumers, making this version more likely to advance than its predecessors.

What makes this different from existing approaches

Legal teeth over voluntary compliance: Unlike NIST's AI Risk Management Framework or company AI principles, this creates enforceable legal requirements with FTC oversight and penalty authority.

Focus on impact assessments, not just documentation: Companies must conduct algorithmic impact assessments that evaluate bias, accuracy, fairness, and privacy implications - going beyond simple disclosure requirements.

Broad definition of covered systems: The bill targets "automated decision systems" that "make or facilitate human decisions" affecting consumers, casting a wider net than narrow AI-specific regulations.

Mandatory external auditing: High-risk systems require independent third-party assessments, preventing companies from simply self-certifying compliance.

Who this resource is for

Legal and compliance teams at companies using automated decision-making systems need to understand potential future obligations and start preparing compliance frameworks.

AI practitioners and data scientists working on systems that affect hiring, lending, healthcare, or other critical decisions should understand how their technical choices might face regulatory scrutiny.

Policy advocates and researchers tracking AI governance developments can use this as a reference for the current state of US federal AI regulation efforts.

Procurement and vendor management teams evaluating AI tools and platforms need to assess whether vendors would meet proposed accountability requirements.

Key compliance hurdles

Defining "high-risk": The bill establishes criteria for high-risk systems but leaves significant interpretation to FTC rulemaking, creating uncertainty for companies trying to assess coverage.

Third-party auditor availability: The requirement for independent assessments assumes a mature market of qualified algorithmic auditors that doesn't fully exist yet.

Technical assessment standards: While the bill mandates impact assessments, it doesn't specify methodologies, potentially leading to inconsistent approaches across companies.

Small business considerations: The legislation includes exemptions for smaller entities, but the thresholds and definitions remain unclear.

Current status and timeline

As of 2023, the bill has been introduced in the Senate but hasn't received committee hearings or markup. Congressional AI policy activity suggests possible movement in 2024, but passage isn't guaranteed. Companies should monitor for:

• Committee hearings and witness testimony
• Industry feedback periods
• Potential amendments addressing business concerns
• House companion legislation
• State-level algorithmic accountability bills that might create compliance complexity

The EU AI Act's implementation timeline offers a preview of how algorithmic governance regulations typically phase in over 2-4 years, giving companies time to prepare compliance systems.