The Great American AI Act of 2026: The Evidence Behind the Federal Push to Preempt State Laws

June 2026 has emerged as the most consequential month for artificial intelligence governance since the technology's commercial breakout. Across three continents, a collision of federal legislation, state-level executive orders, and international summits is replacing theoretical AI ethics with hard legal boundaries. At the center of this shift is the newly drafted Great American Artificial Intelligence Act of 2026, a 269-page bipartisan effort to nationalize AI oversight and preempt a fracturing landscape of state laws.[1][3]

The primary claim driving the federal push is that a patchwork of state-level regulations fundamentally stifles algorithmic innovation. The evidence for this fragmentation is strong. In recent months, Colorado replaced its landmark AI Act with the broader Automated Decision-Making Technology Act, while California, Illinois, and Connecticut advanced their own distinct compliance regimes. The Great American AI Act attempts to resolve this by preempting state laws that specifically regulate AI development for a three-year period, while preserving general state authority over post-deployment activities.[1][3]

However, the evidence supporting the efficacy of federal preemption remains contested. Civil society organizations argue that stalling state laboratories of democracy leaves consumers vulnerable, while industry groups like the Business Software Alliance maintain that unified national standards are the only way to scale frontier models. The draft bill pairs this preemption with mandatory transparency reports, whistleblower protections, and critical safety incident reporting for large-scale developers, mirroring the most stringent state requirements but applying them uniformly.[1][3]

The Great American AI Act's fourth title specifically addresses the evidentiary gap in AI training. It mandates provisions related to public data usage and requires developers to maintain transparent logs of the datasets used to train frontier models. This aligns with growing bipartisan sentiment—highlighted by an Annenberg Public Policy Center survey showing 65 percent of Americans believe the government has done too little to regulate AI—that the black-box nature of model development is no longer acceptable.[1]

A second major claim dominating the June policy landscape is that frontier AI models now constitute critical national security assets requiring strict export controls. The evidence for this shift materialized sharply when the Trump administration issued a directive preventing foreign nationals from accessing Anthropic's newest and most powerful models, Fable 5 and Mythos 5. The administration cited undisclosed security vulnerabilities, effectively treating commercial AI weights as classified munitions.[2]

The uncertainty surrounding this national security framing erupted at the Group of Seven (G7) summit in Evian-les-Bains, France. French President Emmanuel Macron publicly challenged the U.S. directive, labeling it a "strictly nationalist" reaction that threatens global democratic cooperation. Macron argued that cutting off allied access to frontier models could artificially depress the valuations of U.S. tech firms while forcing Europe to aggressively fund sovereign AI alternatives. OpenAI CEO Sam Altman echoed the need for an international forum to draw up guardrails, warning that safety cannot be left solely to the companies building the systems.[2]

Domestically, the claim that AI will cause imminent, large-scale labor displacement has moved from academic theory to executive action. The evidence here is rooted in preemptive state measures rather than historical job-loss data. On May 21, California Governor Gavin Newsom signed Executive Order N-6-26, directing state agencies to draft concrete policies on severance standards, expansions to the Worker Adjustment and Retraining Notification (WARN) Act, and potential revenue-sharing models for AI companies.[4]

Beyond California, the patchwork of employment-related AI laws is already active. Colorado's regulations take effect in June 2026, while Illinois and New Jersey have staggered implementations running through the end of the year. These laws impose distinct requirements for risk assessment and anti-discrimination testing whenever automated decision tools are used in hiring or firing, forcing national employers to adopt the strictest state standard as their baseline.[4]

The uncertainty regarding actual workforce impact is explicitly acknowledged in the California order. Rather than acting on definitive job-loss figures, the state has mandated the creation of an "early warning system" dashboard within 90 days to track AI's sector-by-sector impact on employment. This reflects a broader regulatory consensus: governments lack the empirical data to prove AI's exact economic toll, but the political risk of being caught unprepared has necessitated immediate legislative scaffolding.[4][7]

California's executive order also explores radical new economic models to offset AI's societal costs. By October 2026, state agencies must present options for altering incentive structures, including mandatory programs that would direct a portion of revenue generated by AI companies to support beneficial public deployments. This represents a shift from merely mitigating AI's harms to actively redistributing the financial gains of automation.[4]

In the realm of intellectual property, the claim that AI systems cannot hold copyright has now been cemented by the highest judicial authority in the United States. The evidence is definitive: on March 2, 2026, the U.S. Supreme Court denied certiorari in Thaler v. Perlmutter, ending a years-long legal battle to secure copyright protection for artwork autonomously generated by an AI system named the "Creativity Machine."[6]

By refusing to hear the case, the Supreme Court left intact the D.C. Circuit's ruling that human authorship is a "bedrock requirement" of the Copyright Act. While this provides strong legal certainty that AI is merely a tool rather than an author, it leaves open the complex evidentiary question of exactly how much human prompting, editing, or direction is required to copyright an AI-assisted work. The U.S. Copyright Office maintains that prompts alone function merely as unprotectable ideas, forcing businesses to meticulously document their human creative workflows.[6]

Finally, the claim that the European Union will dictate global AI compliance standards is about to be tested by hard enforcement. The evidence for this global shift is anchored in the August 2026 deadline for the EU AI Act's high-risk system obligations. Systems used in employment, critical infrastructure, and law enforcement must now feature documented risk management, human oversight controls, and conformity assessments before entering the European market.[5]

The extraterritorial reach of the EU AI Act means these rules apply not just to European companies, but to any developer whose AI system affects the EU market. U.S. firms that fail to map their AI coding tools or customer service bots against the EU's Annex III high-risk categories face immediate exposure. Engineering teams are now required to implement traceability and logging mechanisms that were previously considered proprietary internal processes.[3][5]

To operationalize this, EU member states are rapidly standing up domestic enforcement architectures. On June 17, Ireland—home to the European headquarters of major U.S. tech firms—published the Regulation of Artificial Intelligence Bill 2026. The legislation establishes the AI Office of Ireland as an independent statutory body empowered to investigate breaches and levy administrative sanctions.[5]

The stakes for non-compliance are existential for smaller developers and highly punitive for tech giants. The EU framework allows regulators to impose fines of up to €35 million or 7 percent of a company's global turnover for severe infringements. As the U.S. debates the Great American AI Act and the G7 fractures over export controls, the EU's transition from theoretical rule-making to active policing ensures that August 2026 will serve as the true stress test for global algorithmic accountability.[5][7]