WeVote
Log In

Bill

Bill

H 792

An act relating to liability standards for developers and deployers of artificial intelligence systems

2025-2026 Regular Session Introduced by Daisy Berbeco and 17 co-sponsors

Creates a technology-specific liability framework holding AI developers and deployers accountable for safety, with safe harbors for risk management and transparency.

Read first time and referred to the Committee on Commerce and Economic Development
0
WeVote Research Nonpartisan
Bill Summary · H 792

Summary of H.792 (2025-2026) – Vermont

Purpose and intent

H.792 proposes a new framework for liability standards specifically aimed at developers and deployers of artificial intelligence (AI) systems. The bill acknowledges AI systems as products that can shift decision-making away from humans and may pose inherent safety risks. It establishes targeted duties for those who design (developers) and operate or implement (deployers) high-impact and generative AI systems to ensure safety in reasonably foreseeable use and to protect individuals’ rights.

Key definitions

  • Artificial intelligence system: machine-based system with varying autonomy that can generate outputs influencing physical or virtual environments.
  • Generative AI system: AI that can produce derivative content (text, images, video, audio) resembling training data.
  • High-impact AI system: AI used as a controlling factor in consequential decisions, to categorize by protected characteristics, manage critical infrastructure, control vehicles/medical devices, engage in synthetic relationships, or capable of high-performance tasks with significant risk to economic security or health/safety.
  • Deployer: operator of an AI system (internal use or third-party use); excludes individuals or small businesses (defined as ≤19 employees or ≤10,000 product users).
  • Developer: designer/owner/modifier of an AI system.
  • Product and harm concepts: includes specific definitions of product, material facts, warnings, express warranties, and various forms of harm (physical, financial, reputational, emotional, etc.).

Main provisions

  1. Liability framework (developers) – §4193c

    • Developers have limited liability in products liability actions unless the plaintiff proves:
      • (a) defective design caused harm, or
      • (b) inadequate warnings/instructions caused harm, or
      • (c) breach of an express warranty linked to the harm.
    • For design defects, plaintiffs must show:
      • knowledge of danger at release, consideration of intended and foreseeable unintended uses, and
      • existence of a feasible, practical alternative design that would have reduced risk without defeating usefulness.
    • Warning and warranty standards mirror the above causation requirements.
    • A “safe harbor” presumption is available for developers if they:
      • conduct testing/verification/auditing per industry best practices,
      • mitigate risks and disclose them to deployers/consumers,
      • maintain an AI data sheet detailing uses, training data sources, risk management, red-teaming results, and other transparency measures, and
      • meet additional age-related disclosures and responsibilities if the product is likely used by those under 18.

  2. Liability for deployers – §4193d

    • Deployers are treated as liable as developers for harm if they:
      • make material and substantial changes to the product, or
      • misuse the product contrary to express warranties, where such misuse proximately caused harm.
    • If the product is used as intended by the developer, that use is not considered misuse.
    • Licensing the product does not automatically create liability solely due to ownership/use.

  3. Safe harbors for deployers – §4193d(4)-(8)

    • Deployers may enjoy a rebuttable presumption of non-defectiveness if they implement a risk management policy aligned with best practices, tailored to system scope and data processed, and make it accessible to employees and the Attorney General.

  4. Applicability and fault allocation – §4193e

    • Supplements existing tort/product liability law; recognizes joint and several liability between developers and deployers where fault exists.
    • Comparative negligence applies; damages may be reduced based on the plaintiff’s share of fault related to a violation of the AI liability subchapter.

  5. Enforcement – §4193f

    • Violations constitute an unfair and deceptive act in commerce under Vermont law.
    • The Attorney General has authority to adopt rules, investigate, and pursue civil actions or discontinuance agreements.

Effective date

  • The act would take effect on July 1, 2026.

Potential impact

  • Creates a clearer, technology-specific liability regime for AI products, emphasizing safety design, transparent data practices, and risk management.
  • Encourages developers to implement robust testing, documentation, and warnings.
  • Provides deployers with a defined safety framework and potential protections from liability if they adhere to risk-management standards and clearly communicate risks.
  • Establishes a joint-fault framework and explicit enforcement mechanisms to ensure accountability in AI deployment.

Compiled from official sources — confirm details with the bill’s official record.

Sign in to ask a question.