RISE Act of 2025
- Sponsor
- Sen. Lummis, Cynthia M. [R-WY]
- Committees
- Commerce, Science, and Transportation Committee (primary)
- Last action
- Jun 12, 2025
Bottom line
The bill aims to balance AI innovation with accountability by granting developers a safe harbor from liability for AI errors, provided they meet stringent transparency and documentation requirements, thereby shifting some responsibility to the 'learned professional' user.
What it actually does
This bill establishes conditional civil liability immunity for artificial intelligence (AI) developers when their AI products are used by 'learned professionals' (e.g., doctors, lawyers) in providing professional services to clients. To qualify for this immunity, developers must publicly release and continuously maintain 'model cards' and 'model specifications' for their AI products, and provide clear documentation of known limitations and appropriate uses. Immunity does not apply in cases of developer recklessness or willful misconduct, or if transparency documentation is not updated.
Proponents argue
Proponents argue that the bill fosters responsible AI innovation by reducing legal uncertainty for developers, encouraging transparency through required disclosures, and ensuring that beneficial AI tools can be integrated into professional services without stifling development. They emphasize the role of learned professionals in exercising independent judgment and understanding the tools they employ.
Opponents contend
Critics might argue that the bill shifts too much liability away from powerful AI developers onto individual professionals, potentially leaving clients harmed by AI errors with less recourse. They might also contend that the transparency requirements, while helpful, may not be sufficient to fully understand complex AI systems or prevent all foreseeable harms, and that proving 'recklessness' could be difficult.
The bill is relatively short, making it accessible for a quick read, but its legal and technical definitions require careful attention to fully grasp its implications.
Section 1
Short Title
This section establishes the official short title of the Act as the 'Responsible Innovation and Safe Expertise Act of 2025' or the 'RISE Act of 2025.' This is how the bill would commonly be referred to if enacted.
Supporters argue
Supporters argue that a clear and concise short title helps in identifying and referencing the legislation, making it easier for the public and policymakers to discuss and understand its purpose.
Critics contend
No specific opposition is typically raised against short titles, as they are standard legislative practice.
Tradeoffs
None significant for a short title.
Section 2
Congressional Findings on AI and Liability
This section outlines Congress's rationale for the bill, noting the rapid advancement and deployment of AI, the uncertainty regarding liability for AI errors, the need for transparency, and the professional obligations of learned professionals using AI. It emphasizes balancing innovation with transparency and responsibility.
Section 3
Key Definitions for AI Liability
This section defines critical terms used throughout the Act, such as 'artificial intelligence,' 'client,' 'developer,' 'error,' 'learned professional,' 'model card,' and 'model specification.' These definitions are crucial for determining who is covered by the immunity and what conditions apply, particularly the broad scope of 'developer' and the detailed requirements for 'model card' and 'model specification.'
Section 4(a)
Conditional Immunity for AI Developers (Safe Harbor)
This core provision grants civil liability immunity to AI developers for errors generated by their AI products when used by 'learned professionals' in professional services. This immunity is conditional upon the developer publicly releasing and continuously maintaining a 'model card' and 'model specification' for the AI product, and providing clear documentation of its known limitations and appropriate uses.
Section 4(b) and 4(c)
Limitations on Immunity and Developer's Duty to Update
Subsection (b) clarifies that immunity does not apply to acts or omissions constituting recklessness or willful misconduct by the developer. Subsection (c) mandates that developers must update their model cards, model specifications, and documentation within 30 days of deploying a new version or discovering a new material failure mode, or they lose immunity for harms proximately caused by that failure to update.
Section 4(d)
Federal Preemption of State Liability Claims
This subsection establishes that the federal immunity provided by the Act preempts (overrides) any conflicting State law claims against a developer for an AI error, provided the developer qualifies for immunity. However, it explicitly states that claims based on fraud, knowing misrepresentation, or conduct outside the scope of professional use are *not* preempted.
The definition of 'developer' includes not only those who create or modify AI products but also those who '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.'
Section 3(3)
Why it matters:This is standard legislative practice to ensure comprehensive coverage of all relevant actors in the AI supply chain, not necessarily evasive.
Case for: Supporters argue this ensures that all entities presenting an AI product to the public under their brand are accountable for its proper functioning and for meeting the transparency requirements, regardless of whether they built the foundational model. This prevents companies from simply rebranding third-party AI and avoiding responsibility.
Case against: Critics might argue this could place undue burden on distributors or marketers who have limited control over the underlying AI technology, potentially discouraging broader adoption and distribution of AI products due to increased compliance costs.
Estimated impact: Broadens the scope of entities subject to the bill's requirements and potential liabilities/immunities, affecting a wider range of businesses in the AI ecosystem.
Developers are required to release the 'model specification' but may include redactions 'only relating to information that would reveal trade secrets unrelated to the safety of the artificial intelligence product' and only if a written justification for the redaction is furnished.
Section 4(a)(1)(B)
Why it matters:This is a common legislative compromise to balance transparency with intellectual property protection, not necessarily evasive.
Case for: Proponents argue this strikes a necessary balance, allowing developers to protect proprietary information (trade secrets) that doesn't directly bear on the safety or functionality of the AI, while still ensuring critical safety-related information is public.
Case against: Critics might contend that the definition of 'trade secrets unrelated to safety' could be broadly interpreted by developers, potentially allowing them to withhold information that is indirectly relevant to safety or understanding the AI's behavior, thereby undermining the spirit of transparency.
Estimated impact: Influences the actual level of transparency achieved by the bill, potentially limiting public and professional understanding of AI systems.