Artificial intelligence is changing and evolving more quickly than humans realize. The speed of innovation in the private sector has outstripped most of the capacity of federal governments to develop regulations. From generative models capable of producing verisimilar synthetic media to advanced decision-making systems with substantial influence over national security policy, the stakes and risks associated with adversarial AI have never been higher.
Competing with global actors like China for AI supremacy raises the question of whether or not the divided United States can build the necessary guardrails in time, and whether innovation will outrun meaningful regulation before the United States can implement any safeguards.
Federal Frameworks: Building the Foundations
In 2023, President Biden signed Executive Order 14110 to set a national AI policy agenda focused on safety, trust, and equity. The order called on federal regulatory agencies (such as NIST) to create technical standards for AI risk management.
It also directed federal agencies to adopt responsible AI approaches to certain national priority areas such as defense, healthcare, and infrastructure. Executive Order 14110 included transparency, guaranteed safety testing of high-risk systems, and Chief AI Officers in federal departments to ensure that these plans were adopted.
The Biden administration also launched the U.S. AI Safety Institute, which will be responsible for developing benchmarks to evaluate powerful AI and will also coordinate work with allied nations. At the same time, there were voluntary commitments made by some of the major AI companies—OpenAI, Google, and Microsoft—that offered the adoption of some safety assurance strategies, such as requiring external audits, watermarking AI-generated output, and recommending disclosures about system capabilities.
Although actions and commitments, whether voluntary or mandated, are a meaningful first step in the short term, they are all contingent upon executive action; this has implications for the long-term sustainability of the plans.
Shifting Course: The 2025 Pivot
With the new administration in 2025, Trump revoked Executive Order 14110 and presented a new policy approach aimed at promoting AI innovation unimpeded by regulation. Executive Order 14179, "America First in AI," called for expedited deployment, competitive tensions with the economy, and, relevantly, a much lighter touch on national defense, while seriously rolling back the earlier ethics and safety-oriented policy.
This change represented a dramatic shift from alignment and ethics to acceleration. The AI Action Plan emphasized expanding domestic AI capacity, building an infrastructure of public-private partnerships to deploy models rapidly, and deregulating to counter the race for superiority with rivals like China.
Supporters of this shift lauded these changes as essential to preserving global leadership, whereas critics worried it would leave regulatory vacuums around safety, bias mitigation, and catastrophic prevention.
State Action: Laboratories of AI Democracy
Without federal law, states have become laboratories for AI regulation. For instance, California’s SB-1047 was introduced as a comprehensive framework for regulating high-computational frontier models, including licensing, compliance audits, and risk disclosures for national security or whistleblowers.
The bill ultimately got vetoed, but it did highlight an interest in state-led proactive regulation. Other states, as shown with AI-generated deepfakes, biometric data usage, and chatbot clarity, have also explored legislation.
Regrettably, this hybrid of rules has also resulted in ambiguity and uneven implementation, prompting industry bodies to demand a comprehensive strategy. The present administration has proposed ways to, in effect, preempt state laws, but such plans remain controversial in many state legislatures that wish to maintain their flexibility to provide responsive and local regulations.
The Legislative Gap
Although Congress has introduced a number of bills related to AI—including the CREATE AI Act, designed to expand AI research infrastructure, and the Global Catastrophic Risk Mitigation Act, addressing long-term alignment risks—no comprehensive legislation has thus far been enacted.
Partisan divides, the speed of shifting technology, and intense lobbying by corporations have complicated any legislative process. Currently, the agencies that are tasked with regulatory enforcement are restricted by statutory authority in what they can and cannot enforce. Much of what the agencies can enforce is a framework built on voluntary compliance and executive action that can and does change every time there is a change in administration.
Without legislation from Congress providing authority and oversight through regulation of AI tech development and deployment, the country will suffer in building durable regulatory institutions.
Racing the Clock: Innovation vs. Safety
The federal pace of policymaking is still significantly slower than that of those in the private sector, where development is rapidly increasing. New model architectures, pipelines, and emergent capabilities are being developed every month, showing how quickly even good-faith legislation could become outdated.
While early industry insiders were calling for regulation of AI, many now say that overregulation could stifle gold-plated innovation—innovation that may be lost to less regulated nations abroad. This innovation-first approach is particularly prominent within U.S.-China competition. Due to the growing perception of artificial intelligence (AI) as a key platform of geopolitical power, an increasing number of the world's policymakers have taken the position that states should seek to maximize national advantages over global coordination and long-term safety.
However, this zero-sum game could disrupt the more collaborative global work toward lessening the harm associated with AI systems created by humans and possibly operated beyond the control of humans.
Bridging the Gap: What’s Needed
If the U.S. is to close the rapidly widening gap between technology and regulation, the technology governance should be layered. To put that more formally, a federal regulatory approach that requires minimum standards for evaluation, testing, and disclosure will be an important first step to allow states the flexibility to develop regulations that are effective locally. Entities with public trust, like the AI Safety Institute, need the appropriate funding and regulatory authority to set the minimum standards and enforce them.
Moderate threshold-based oversight of only the most powerful models could help strike a more focused, less burdensome alternative to broad top-down regulatory guidelines. The criteria for oversight would require that developers of frontier models who meet the regulatory threshold undergo an external audit for public transparency, register their models' capabilities in a public repository, and submit post-deployment impact reports to outline the societal consequences of their systems.
While this would help align the innovation agenda with accountability at the frontier, Congress must take action. Only through accomplished bipartisan legislation can the U.S. build durable governance mechanisms in the space of AI, regardless of executive mandates or private-sector resistance.
Only a Matter of Time
The U.S. is at a pivotal moment in the dawn of the AI age. Only a few years ago, when the Biden administration took office, the U.S. acted sequentially to build the framework for AI safety and alignment. In a short period of time, the political environment shifted toward prioritizing economic and strategic acceleration, deprioritizing executive caution.
While states experiment, Congress stalls, and the private sector races ahead, the U.S. must endeavor not only to lead in innovation but also to lead in development with the same level of accountability. While the stakes of governance and competition may appear limited to national interests, the competitive landscape speaks to a larger issue of human-AI futures.
The window is narrowing for the U.S.'s efforts to secure leadership in AI. It is unknown when it will be too late for effective governance of AI to take place.