Purpose

This bill creates a formal framework in Minnesota to regulate artificial intelligence (AI) safety verification. It establishes a licensing system for Independent Verification Organizations (IVOs) to assess AI models and AI applications for risks that could cause personal injury or property damage. It also creates an AI advisory council to oversee activities and adds a legal assumption related to liability in AI-related injury cases.

Main Provisions

• Definitions and scope

  • Defines key terms: Artificial intelligence application, Artificial intelligence model, Deployer (the entity implementing an AI in Minnesota), Developer (the entity creating an AI for deployment in Minnesota), Independent verification organization or IVO (licensed to assess AI safety), Security vendor (third party assisting with safety analysis), and Commissioner (Minnesota’s commissioner of commerce).

• IVO licensure and plan requirements

  • Requires an IVO to obtain a license from the commissioner and to file a detailed plan.
  • The plan must identify risks that AI models or applications must mitigate, define acceptable risk levels, specify measurable metrics and target levels, describe data sources and methods for measurement, and outline an ongoing evaluation and reporting protocol.
  • The plan must include technical, operational, and governance requirements for developers or deployers (predevelopment and postdevelopment risk mitigation, ongoing monitoring, risk assessment methodologies, audit approaches, and whistleblower protections/training).

• License determination and scope

  • The commissioner may license an IVO if the plan demonstrates independence from the AI industry and adequately mitigates identified risks.
  • If a plan only covers certain risks, the license can limit verification to those risks and markets.
  • The license specifies which risks and markets the IVO is authorized to verify.

• License enforcement and changes

  • The commissioner may revoke an IVO license for misrepresentation, failure to adhere to the plan, material harm, or if changes render the verification methods obsolete.
  • There is a cure process allowing a licensed IVO to correct issues before license termination.
  • The commissioner can establish reasonable application and renewal fees to cover administrative costs.
  • The commissioner may adopt rules to implement these sections.

• IVO implementation, updates, and reporting

  • Licensed IVOs must implement their approved plan and revoke verification if developers or deployers fail to meet mitigation requirements, fail monitoring, or violate governance policies.
  • IV0s may update or modify plans, evaluation benchmarks, audit methodologies, governance plans, and verification activities to adapt to new technology; material changes must be reported to the commissioner with rationale and effect.
  • Annual reporting to the commissioner and to legislative committees about AI capabilities, societal risks and benefits, resource adequacy, verification results, remediation activities, and governance/funding changes; redacted or anonymized information may be used as appropriate; records must be retained for ten years and the commissioner will publish redacted reports.

• AI Advisory Council

  • Establishes an Artificial Intelligence Advisory Council within the Department of Commerce.
  • Council duties include delegated licensing and auditing authority; members must be free of conflicts of interest, refrain from incompatible employment or ownership interests in AI companies, and observe a one-year post-employment restriction with AI firms or IV0s.
  • The council’s administration includes terms, reimbursement for expenses, potential compensation, removal for misconduct, and a quorum requirement for decisions.
  • The council must maintain records relating to licensing decisions.

• Liability presumption

  • Creates a rebuttable presumption of non-liability in civil actions for personal injury or property damage caused by an AI model or AI application if: 1) the model/application was verified by a licensed IVO at the time of injury, 2) the injury arose from a risk the IVO was licensed to verify, and 3) the model/application falls within the market segment for which the IVO was licensed.

Significant Changes to Existing Law

• New regulatory framework: Establishes a state-regulated licensing regime for AI safety verification (IVOs), which did not previously exist in Minnesota.
• New governance body: Creates the Artificial Intelligence Advisory Council to oversee licensing, auditing, and related governance activities.
• Expanded accountability and transparency: Requires ongoing reporting, plan updates, and publication of redacted IVO reports by the Department of Commerce.
• Liability policy shift: Introduces a statutory rebuttable presumption limiting liability in certain AI-related injuries when the AI was verified by a licensed IVO and within its verified risk scope.
• Risk-based verification: Empowers the commissioner to license IVOs for specific risks and markets, allowing partial verification scopes rather than a blanket approval.
• Whistleblower protections and governance emphasis: Embeds protections and governance expectations for developers, deployers, and IVO staff within risk management requirements.

Relevant terms, processes, and governance concepts are embedded throughout these provisions, including ongoing monitoring, corrective actions, governance plans, data sources, target risk levels, and annual reporting to lawmakers.

Relevant Terms - Artificial intelligence (AI) - Artificial intelligence application - Artificial intelligence model - Independent verification organization (IVO) - License / licensure - Commissioner (Minnesota Commerce Commissioner) - Deployer - Developer - Security vendor - Risk mitigation - Governance plan - Evaluation benchmarks - Audit methodologies - Ongoing monitoring - Corrective actions - Whistleblower protections - Advisory Council / Artificial Intelligence Advisory Council - Market segment - Verification reports (and redacted publication) - Annual reporting - Cure opportunity - Limited liability rebuttable presumption - Rulemaking

If you’d like, I can tailor this summary to emphasize how it would affect developers, deployers, or consumers, or simplify further for a specific audience.