VET Artificial Intelligence Act

Bill Number

S. 2615

Origin Chamber

Senate

Congress

119th Congress, Session 1

Policy Area

Science, Technology, Communications

Status

Introduced

Latest Action

2025-07-31: Read twice and referred to the Committee on Commerce, Science, and Transportation.

Last Updated

2026-02-27T12:03:21Z

AI-Generated Summary

Purpose

The Validation and Evaluation for Trustworthy (VET) Artificial Intelligence Act aims to create voluntary, evidence-based guidelines to build trust in artificial intelligence (AI) systems. It focuses on testing, evaluating, validating, and verifying AI to ensure accountability and governance, while aligning with the National Institute of Standards and Technology's (NIST) existing AI Risk Management Framework. The goal is to encourage safe AI adoption by addressing risks based on how the AI is used and its potential impacts.

Key Provisions

• Definitions (Section 3): Establishes clear terms, such as:
• AI system: A machine-based tool that processes inputs to produce outputs like predictions or decisions affecting real or virtual environments.
• Developer: An entity that creates or owns an AI system (not just uses it).
• Deployer: An entity that operates an AI system for itself or others.
• Internal AI assurance: An impartial evaluation done internally by the developer or deployer to check functionality, risks, and governance.
• External AI assurance: An independent review by a nonaffiliated third party (unrelated to the developer/deployer, financially independent, and expert in AI risks like data privacy) to verify claims and identify issues like vulnerabilities or societal harms.

• Voluntary Guidelines Development (Section 4): Within one year of enactment, NIST's Director must collaborate with public and private groups (e.g., National Science Foundation, Department of Energy) to create and update every two years a set of voluntary technical guidelines for internal and external AI assurances. These must cover:
• Standards for privacy protections, harm mitigation (e.g., bias or errors affecting people), data quality, documentation/transparency, and governance controls.
• Best practices for testing methods, deciding when/how often to conduct assurances (based on AI risk level), scoping evaluations, disclosing results, and taking corrective actions.
• Alignment with international and industry standards, without mandating specific tools or technologies.
• Protections for confidential data during evaluations.
• NIST must seek public input via comments, drafts, workshops, and forums with diverse stakeholders before publishing the guidelines online.

• Advisory Committee (Section 5): Within 90 days of guidelines publication, the Secretary of Commerce establishes a temporary Artificial Intelligence Assurance Qualifications Advisory Committee (up to 20 experts from academia, industry, consumers, civil rights, labor, etc.). It reviews case studies on certifications and recommends:
• Qualifications (e.g., expertise, independence, licensing) for those conducting assurances.
• Ways to use existing programs for accreditation.
• The committee submits a public report to Congress and the Secretary within one year and then dissolves.

• Study and Report on Assurance Entities (Section 6): Within 90 days of guidelines publication, the Secretary of Commerce studies organizations providing AI assurances, assessing their capabilities (e.g., staff, tools, infrastructure), data protection practices, market demand, and potential use of existing accredited labs. A report with recommendations goes to Congress and relevant agencies within one year.

Significant Changes to Existing Law

This bill introduces new requirements for NIST to develop and maintain specific voluntary guidelines for AI assurances, building directly on but expanding the scope of NIST's 2023 AI Risk Management Framework (and successors). It does not amend prior laws but creates fresh mechanisms like the advisory committee and sector study, emphasizing voluntary standards over mandatory regulations. No direct repeals or overrides of existing AI-related laws (e.g., those on data privacy) are included.

Potential Impacts

• Government Agencies: NIST and the Department of Commerce gain responsibilities for guideline development, stakeholder engagement, and reporting, potentially increasing workload and coordination with other agencies like Energy and NSF. This could standardize federal approaches to AI evaluation without new enforcement powers.
• Citizens: Enhances trust in AI by promoting evaluations that address privacy, harms (e.g., discrimination or errors), and transparency, potentially reducing risks in everyday AI uses like recommendations or decisions in hiring/healthcare. No direct mandates, so benefits depend on voluntary adoption.
• Businesses and AI Sector: Developers and deployers may face indirect pressures to adopt guidelines for credibility and market access, fostering innovation while encouraging better risk management. Third-party evaluators could see growth in demand for services.
• International Relations: By aligning with global standards, the U.S. could lead in trustworthy AI, influencing international norms and trade (e.g., easier exports of compliant AI). It avoids prescriptive rules that might conflict with foreign regulations.

Main Stakeholders Affected

• AI Developers and Deployers: Primary targets, as they must consider guidelines for internal evaluations and decide on external ones based on AI risks.
• Third-Party Assurance Providers: Independent evaluators (nonaffiliated experts) who conduct external reviews; the bill studies and supports their growth.
• Consumers and Advocacy Groups: Benefit from protections against AI harms; represented in stakeholder input and the advisory committee.
• Government and Research Entities: NIST, Commerce Department, NSF, and Energy Department lead implementation; academia and labs provide expertise.
• Civil Society and Labor: Civil rights, public health/safety, workforce, and labor organizations influence guidelines via consultations and committee roles, ensuring broader societal considerations like equity and job impacts.
• Industry and Accreditation Bodies: AI companies, professional certifiers, and nonprofits help shape standards and may expand services.

Notable Legal, Constitutional, or Political Implications

• Legal: The voluntary nature avoids regulatory burdens, respecting innovation and avoiding First Amendment issues around mandated disclosures. It promotes accountability through best practices rather than penalties, potentially serving as a foundation for future laws if adoption is low.
• Constitutional: No direct conflicts; emphasizes impartiality and independence in evaluations, aligning with due process principles for fair AI assessments without government overreach.
• Political: Bipartisan sponsorship (Sens. Hickenlooper and Capito) signals broad support for non-regulatory AI governance. It positions the U.S. as proactive on AI safety amid global competition (e.g., EU's stricter rules), but critics might view it as insufficiently binding, while supporters see it as balanced for economic growth. Referred to Senate Commerce Committee, it could influence broader AI policy debates.

This summary was generated by AI and may contain inaccuracies. Refer to the official source document for the authoritative text.

Sponsor

Sen. Hickenlooper, John W. [D-CO]

Cosponsors (3)

Sen. Capito, Shelley Moore [R-WV], Sen. Slotkin, Elissa [D-MI], Sen. Young, Todd [R-IN]

Recent Actions

• 2025-07-31: Read twice and referred to the Committee on Commerce, Science, and Transportation.
• 2025-07-31: Introduced in Senate

Bill Versions

• Validation and Evaluation for Trustworthy (VET) Artificial Intelligence Act — issued 2025-07-31 — PDF (15 pages)