RISE Act of 2025

Summary

Establishes conditional immunity from civil liability for AI developers when their systems are used by learned professionals in professional services, provided specific transparency and documentation criteria are met.

Requires developers to publicly release and maintain a model card and model specification before AI deployment, allowing limited redactions for trade secrets with justifications.

Mandates developers to provide documentation detailing AI systems' limitations and appropriate use domains.

Specifies that immunity does not cover acts of recklessness or willful misconduct and excludes claims for fraud or misrepresentation.

Includes a duty for developers to update relevant documentation within 30 days of new AI versions or discovered failure modes.

Expressly preempts state law claims except those based on conduct outside professional use or on fraud and misrepresentation.

Takes effect on December 1, 2025, and applies to acts or omissions from that date onward.

Establishes immunity from civil liability for certain artificial intelligence developers.

A BILL To establish immunity from civil liability for certain artificial intelligence developers, and for other purposes.

Be it enacted by the Senate and House of Representatives of the United States of America in Congress assembled,

SECTION 1. Short title.

This Act may be cited as the “Responsible Innovation and Safe Expertise Act of 2025” or the “RISE Act of 2025”.

Acknowledges AI advancements, liability uncertainties, transparency issues, and emphasizes the need for clear AI governance standards.

SEC. 2. Findings.

Congress finds the following:

(1) Artificial intelligence systems have rapidly advanced in capability and are increasingly being deployed across professional services, including healthcare, law, finance, and other sectors critical to the economy.

(2) Industry leaders have publicly acknowledged the development of increasingly powerful artificial intelligence systems, with some discussing the potential for artificial general intelligence and superintelligence that could fundamentally reshape the society of the United States.

(3) The current lack of clarity regarding liability for artificial intelligence errors creates uncertainty that impedes the responsible integration of these beneficial technologies into professional services and economic activity.

(4) Many artificial intelligence systems operate with limited transparency regarding their capabilities, limitations, and default instructions, making it difficult for professional users to assess appropriate use cases and for legal systems to fairly allocate responsibility when errors occur.

(5) Learned professionals who utilize artificial intelligence tools in serving clients have professional obligations to understand the capabilities and limitations of the tools they employ, requiring access to clear information about system specifications and performance characteristics.

(6) Establishing clear standards for artificial intelligence transparency, coupled with appropriate liability frameworks, will promote responsible innovation while ensuring that the benefits and risks of artificial intelligence systems are properly understood and managed as these technologies continue to advance.

(7) The development of artificial intelligence systems that may significantly impact the future of human civilization warrants a governance approach that balances innovation incentives with robust transparency requirements and appropriate allocation of responsibility among developers, professional users, and other stakeholders.

Defines key terms: artificial intelligence, client, developer, error, learned professional, model card, for AI governance.

SEC. 3. Definitions.

In this Act:

(1) ARTIFICIAL INTELLIGENCE.—The term “artificial intelligence” has the meaning given the term in section 5002 of the National Artificial Intelligence Initiative Act of 2020 (15 U.S.C. 9401).

(2) CLIENT.—The term “client” means a person that— (A) engages the services of a learned professional; (B) relies upon the expertise, judgment, and advice of the learned professional; and (C) has a relationship with the learned professional that is governed by professional standards, codes of conduct, or regulations.

(3) DEVELOPER.—The term “developer” means a person that— (A) creates, designs, programs, trains, modifies, or substantially contributes to the creation or modification of an artificial intelligence product; (B) exercises control over the design specifications, functionality, capabilities, limitations, or intended uses of an artificial intelligence product; or (C) markets, distributes, licenses, or makes available an artificial intelligence product under their own name, brand, or trademark, regardless of whether the person creates the original underlying technology of the artificial intelligence product.

(4) ERROR.—The term “error” means— (A) any output, action, recommendation, or material omission by an artificial intelligence product that is false, misleading, fabricated, deceptive, or incomplete in a manner that a reasonable developer could foresee would cause harm; or (B) any failure of an artificial intelligence product to perform a function or task that the artificial intelligence product expressly or implicitly represents itself as capable of performing.

(5) LEARNED PROFESSIONAL.—The term “learned professional” means an individual who— (A) possesses specialized education, training, knowledge, or skill in a profession; (B) is licensed, certified, or otherwise authorized by an appropriate Federal or State authority to practice in that profession; (C) is bound by professional standards, ethical obligations, and a duty of care to clients; and (D) exercises independent professional judgment when using tools, including artificial intelligence products, in the course of rendering professional services.

(6) MODEL CARD.—The term “model card” means a publicly available technical document in which a developer describes, consistent with industry standards and as rigorously as or more rigorously than industry peers, the training data sources, evaluation methodology, performance metrics, intended uses, limitations, and risk mitigations, including detection, evaluation, management, and safeguards against errors, of an artificial intelligence product.

Defines "model specification" as configuration instructions influencing AI product outputs, including system prompts and training materials.

(7) MODEL SPECIFICATION.—The term “model specification”— (A) means the text or other configuration instructions of an artificial intelligence product— (i) supplied by a developer; (ii) that establish the intended base behavior, tone, constraints, or goals of the artificial intelligence product; and (iii) that materially influence the outputs of the artificial intelligence product across users or sessions, including the system prompt provided to the model before engaging with user queries; and

(B) includes— (i) the system prompt and any other text or images that the artificial intelligence product receives that are not visible to the end user; (ii) any constitution or analogous guiding document used when training or fine‑tuning of an artificial intelligence product, including in automated schemes in which an artificial intelligence system trains another artificial intelligence system; and (iii) the instructions, rubrics, or other guidance provided to human raters or evaluators of an artificial intelligence product the feedback of whom is used to train or fine-tune the artificial intelligence product.

Grants AI developers immunity from civil liability if they meet documentation and transparency requirements and avoid misconduct.

SEC. 4. Conditional immunity from civil liability for artificial intelligence developers.

(a) Safe harbor eligibility.—A developer shall be immune from civil liability for errors generated by an artificial intelligence product when used by a learned professional in the course of providing professional services to a client if the developer—

(1) prior to deployment of the artificial intelligence product, publicly releases and continuously maintains— (A) the model card for the artificial intelligence product; and (B) the model specification for the artificial intelligence product, which may include redactions— (i) only relating to information that would reveal trade secrets unrelated to the safety of the artificial intelligence product; and (ii) only if the developer furnishes contemporaneously with each redaction a written justification for the redaction identifying the basis for withholding the information as a trade secret; and

(2) provides clear and conspicuous documentation to learned professionals describing the known limitations, failure modes, and appropriate domains of use for the artificial intelligence product.

(b) Scope of immunity.—The immunity provided under subsection (a) shall be conferred to a developer only for acts or omissions that do not constitute recklessness or willful misconduct by the developer.

Prohibits AI developer immunity if model card updates aren't made within 30 days of deployment or discovery.

(c) Duty To update.—Immunity under subsection (a) relating to an artificial intelligence product shall not apply to a developer— (1) that does not update the model card, model specification, and documentation with respect to the artificial intelligence product as described in subsection (a)(1) by the date that is 30 days after the date on which the developer— (A) deploys a new version of the artificial intelligence product; or (B) discovers a new and material failure mode affecting the artificial intelligence product; and (2) of which the failure to make an update described in paragraph (1) by the applicable date described in that paragraph proximately causes a harm occurring after that date.

(d) Preemption.— (1) EXPRESS PREEMPTION.—This section shall apply to any claim arising under State law against a developer for an error arising from the use of an artificial intelligence product by a learned professional in providing professional services if the developer is immune from civil liability under subsection (a). (2) CLAIMS NOT PREEMPTED.—Nothing in this section shall apply to a claim arising under State law against a developer based on fraud, knowing misrepresentation, or conduct outside the scope of professional use of an artificial intelligence product by a learned professional.

Clarifies the Act's effect date and applicability to acts or omissions after December 1, 2025.

SEC. 5. Preservation of other immunities and privileges.

Nothing in this Act shall be construed to affect any immunity from civil liability established by Federal or State law or available at common law that is not related to the immunity established under section 4(a).

SEC. 6. Effective date; applicability.

This Act—

(1) shall take effect on December 1, 2025; and

(2) shall apply to acts or omissions occurring on or after the date described in paragraph (1).