Repeated efforts by Congress to enact comprehensive AI regulations have tripped up over one political obstacle or another. But the recently circulated discussion draft of the Great American Artificial Intelligence Act (GAAIA) may have found a route around the biggest roadblocks according to a recent analysis by Taft Stettinius & Hollister.
Unlike previous federal technology bills that have struggled over broad preemption and sweeping mandates, the nearly 270-page proposal takes a narrower approach by focusing primarily on the developers of powerful frontier AI models while leaving most AI deployers outside its direct regulatory scope.
Released June 4 by Reps. Jay Obernolte (R-CA), and Lori Trahan, (D-MA) the discussion draft reflects growing bipartisan interest in establishing a federal framework for AI governance amid an increasingly fragmented state regulatory landscape. According to the Taft analysis, the bill is designed to balance innovation, transparency and security while avoiding some of the political obstacles that have stalled federal privacy legislation.
A central feature of the proposal is its focus on so-called frontier AI models, defined broadly as highly capable general-purpose systems trained on extensive datasets using substantial computing resources. The bill would impose its most significant obligations on organizations with annual revenue exceeding $500 million that develop these foundational models, effectively targeting companies such as OpenAI, Anthropic and other leading AI developers rather than businesses that simply deploy AI tools.
Under the draft legislation, covered developers would be required to publish AI governance frameworks and regular transparency reports detailing model capabilities, risk assessments, security controls and oversight procedures. They also would be required to report “critical safety incidents” to federal authorities, including unauthorized access to model weights, failures of risk-mitigation measures or situations involving loss of control over advanced systems.
Related: Rhode Island Approves New Rules on Lawyer Use of AI
To oversee those requirements, the legislation would establish a new Center for AI Standards and Innovation within the Department of Commerce. The center would develop voluntary guidelines and security standards, support the work of the National Institute of Standards and Technology (NIST), and monitor domestic and foreign AI developments. The proposal also would authorize the center to license Independent Verification Organizations responsible for auditing frontier model developers and evaluating compliance with AI risk-management frameworks.
Another key difference from previous federal efforts is the GAAIA’s narrow preemption of state AI regulations. Rather than broadly displacing state rules, the discussion draft would preempt state laws governing the development of AI models for only three years after enactment. The narrow and temporary preemption provision appears aimed at addressing one of the most contentious issues in AI policy, per Taft: whether Congress should override a growing patchwork of state AI laws. By limiting preemption to model development and imposing a sunset period, lawmakers appear to be seeking a compromise between federal uniformity and state regulatory authority.
Beyond frontier-model governance, the proposal includes provisions addressing cybersecurity, workforce development and research. It would establish whistleblower protections for employees who report violations of federal AI law, increase penalties for AI-enabled fraud and impersonation of government officials, expand federal support for AI education and research, and require greater transparency regarding certain AI-related layoffs. The bill also calls for international cooperation on technical AI standards.
The Taft analysis notes that while the direct compliance burden would fall largely on major model developers, the effects could spread throughout the AI ecosystem as compliance obligations flow into vendor contracts and procurement requirements. As a result, even companies that are not directly regulated may encounter new expectations from customers and business partners.
Although the legislation has not yet been formally introduced and already faces opposition, the proposal offers one of the clearest indications yet of how Congress may seek to regulate frontier AI at the federal level.
