On June 4, Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA) released a sweeping discussion draft of their Great American Artificial Intelligence Act. The latest bipartisan AI legislation quickly met bipartisan skepticism, particularly concerning the draft’s approach to federal preemption of state AI rules, with many House Democrats opposing the broad preemption for frontier model developers, while many House Republicans and other stakeholders lamented the bill’s omission of preemption for state laws reaching other parts of the AI ecosystem. 

The bill would also establish mandatory disclosure and risk-mitigation requirements for frontier models and task the Center for Artificial Intelligence Standards and Innovation (CAISI) at the National Institute of Standards and Technology (NIST) with oversight of federal AI-related research and analysis, standards and guidelines development, and risk-mitigation activities.  

Federal Preemption

In contrast with earlier AI preemption legislation, the Obernolte-Trahan draft would preempt state laws “specifically regulating the development of any artificial intelligence model” for three years, while expressly declining to preempt state laws and regulations “applicable to activities occurring upon or after the deployment of an artificial intelligence model.” If the bill were enacted, states could no longer regulate the development of AI models but could continue to regulate the deployment and use of AI models and applications downstream, such as companion chatbots. The bill also would not preempt any state law or regulation of general applicability.

This preemption language is narrower than the sweeping legislation that Congress failed to advance last year as part of the budget reconciliation package known as the “One Big Beautiful Bill Act.”  The House had considered a ten-year “moratorium” on enforcing state and local rules to regulate AI.  Thereafter, Senator Ted Cruz (R-TX) offered revised language to impose a “pause” on AI rules as a condition of receiving additional Broadband Equity, Access, and Deployment (BEAD) funding; the Senate ultimately rejected this proposal 99-1, and the bill was enacted without any AI preemption or moratorium provision.

The draft also targets fewer state laws than President Trump’s December 2025 executive order, which orders the Justice Department to sue to invalidate state AI laws that run counter to Administration policy. 

However, even the more limited preemption language remains a nonstarter for some House Democrats who have expressed concern about the rapid development of the most capable AI models while Congress struggles to reach consensus on AI regulation.  Rep. Ted Lieu (D-CA), who has introduced his own comprehensive AI bill with Rep. Obernolte, criticized the proposal as failing to address the concerns of civil rights groups, labor organizations, and watchdog groups.  It has also drawn significant backlash from state lawmakers.  Meanwhile, industry groups have been slow to lend their support to the discussion draft, even as the state regulatory patchwork continues to expand in the absence of comprehensive federal rules.  The White House, on the other hand, supports more fulsome preemption and reportedly favors prioritizing preemption of state AI laws that focus on consumer protection.  

Frontier Model Developer Requirements

The bill’s requirements for frontier model developers generally mirror similar laws in California, Connecticut, and New York, as well as the recently passed SB 315 in Illinois. The bill would apply to “frontier developers” (defined as entities who train, or initiate the training of, a frontier model using computing power greater than 10^26 integer or floating point operations) and “large frontier developers” (defined as frontier developers with annual gross revenue over $500M in the preceding year) and requires entities to satisfy detailed public safety disclosure and reporting requirements. These requirements include:

• Frontier AI Frameworks: Large frontier model developers must write, implement and disclose on a publicly available website a frontier AI framework for each of their frontier models that “includes information relating to how the developer carries out” catastrophic risk identification, assessment, mitigation, and third-party evaluation. The framework must also include information on how the developers address cybersecurity threats and critical safety incidents, as well as their internal governance practices to ensure the framework is implemented. The developer must review each framework at least once per year, and, where there are material updates, clearly and conspicuously publish the modified framework and a justification within 30 days.

• Transparency Reports: Frontier model developers must, before or concurrently with deploying a new or substantially modified frontier model, publish on a publicly available website a transparency report that includes: a mechanism that enables an individual to communicate with the frontier developer; languages supported by the model; “output modalit[ies]” (i.e., format or means of transmitting information); the developer’s intended use cases for the model; any generally applicable restrictions or conditions on uses of the model; and any assessments the developer has conducted concerning potential catastrophic risks arising from use of the frontier model, including any third parties involved in the assessment.

• Third Party Audits: Notably, the bill includes a third-party audit requirement similar to that imposed by SB 315 in Illinois.  The draft would require large frontier model developers to semi-annually retain a CAISI-licensed independent verification organization tobothverify compliance with their own frontier AI frameworks and assess the adequacy of the framework and other policies for mitigating catastrophic risks.

• Reporting: CAISI must establish a mechanism that the frontier developer or member of the public can use to report a critical safety incident, and frontier developers must report a critical safety incident to CAISI within 15 days of discovery. Frontier developers must also report a critical safety incident that poses an imminent risk of death or serious physical injury to a law enforcement agency with relevant jurisdiction within 24 hours of discovery.

Representatives Obernolte and Trahan have not yet formally introduced their discussion draft.  While Congress could conceivably move quickly to consider the bill or any other AI legislation, in light of bipartisan disagreement over the core issue of federal preemption and the dwindling number of legislative days remaining in the 119th Congress, it is unlikely the Great American AI Act will advance as drafted this year.  However, as more advanced AI models become available and states continue to enact AI regulations despite the threat of federal litigation, members of Congress from both parties may feel increased pressure to find a consensus approach on preemption.