U.S. Considers Mandatory A.I. Model Checks by 2025

By VirentaNews Staff — May 10, 2026

More than 70% of the world’s most powerful artificial intelligence models were developed in the United States, according to a 2023 Stanford Institute for Human-Centered AI report, yet none are currently subject to mandatory pre-release safety evaluations. This regulatory gap is now under urgent review by the Biden administration, which is drafting proposals to require government vetting of high-capability AI systems before they enter public or commercial use. The move marks a stark departure from the hands-off approach favored during the Trump years and reflects rising alarm over AI’s potential to spread disinformation, enable autonomous weapons, or destabilize financial systems. With models like GPT-4 and Gemini demonstrating near-human reasoning, the White House is racing to establish guardrails before a crisis forces reactive legislation.

A Strategic Pivot in National Tech Policy

The shift toward pre-deployment oversight represents a fundamental rethinking of how the U.S. governs emerging technologies. For decades, American innovation policy has prioritized speed and market-driven development, particularly in software and digital platforms. But AI’s dual-use nature—its ability to power medical breakthroughs or enable mass surveillance—has prompted a reassessment. The National Institute of Standards and Technology (NIST) has already laid groundwork with its AI Risk Management Framework, released in 2023, which recommends voluntary assessments for bias, security, and transparency. However, voluntary standards lack enforcement, and incidents like AI-generated robocalls mimicking President Biden’s voice during the 2024 primaries have underscored the need for binding rules. The White House now sees proactive oversight as essential to maintaining public trust and global leadership in responsible AI.

Scope and Mechanism of Proposed Vetting

The emerging framework would likely target foundation models exceeding specific performance or computational thresholds—systems trained on vast datasets capable of performing a wide range of tasks. According to draft documents reviewed by Reuters, the Department of Commerce and AI Safety Institute could be tasked with evaluating models for risks related to national security, privacy, and systemic harm. Developers might be required to submit technical documentation, undergo third-party audits, and demonstrate mitigation strategies for misuse. The process could mirror aspects of the FDA’s approval system for medical devices or the FAA’s certification of aircraft. While not all AI systems would fall under this regime, the largest models from firms like OpenAI, Google, and Anthropic would likely be subject to scrutiny, particularly if deployed in sensitive domains like law enforcement or critical infrastructure.

Drivers Behind the Regulatory Push

The push for pre-release vetting has been fueled by a confluence of technological acceleration and geopolitical competition. In 2023 alone, the number of AI models with more than 100 billion parameters tripled, according to the AI Index at Stanford University. At the same time, adversarial actors have demonstrated increasing ability to weaponize AI: deepfake videos have disrupted elections in Nigeria and Slovakia, while automated disinformation campaigns have targeted U.S. military personnel. Internationally, the European Union has already passed the AI Act, which includes strict requirements for high-risk systems, and China has implemented rules on algorithmic transparency. U.S. officials fear falling behind in setting global norms. As National Security Advisor Jake Sullivan stated in a 2023 speech, “The nation that sets the rules for AI will shape the trajectory of the 21st century.” This geopolitical calculus, combined with domestic pressure from civil society and bipartisan lawmakers, has made regulation politically viable.

Implications for Industry and Innovation

While intended to enhance safety, mandatory vetting could slow deployment timelines and increase compliance costs, particularly for startups lacking legal and technical resources. Critics warn that overregulation might push innovation overseas or entrench dominance among well-resourced tech giants who can navigate complex approval processes. Conversely, proponents argue that clear, predictable rules could reduce long-term liability and foster public adoption. Sectors like healthcare and autonomous transportation may benefit from standardized validation, accelerating integration of trustworthy AI. Ultimately, the design of the framework will determine whether it becomes a burden or a foundation for sustainable growth. The balance between agility and accountability will define America’s competitive edge in the AI era.

Expert Perspectives

Opinions among technologists and policy experts remain divided. Fei-Fei Li, co-director of Stanford’s Human-Centered AI Institute, supports vetting for models with broad societal impact, stating, “With great capability comes great responsibility.” Others, like former OpenAI researcher Miles Brundage, caution against one-size-fits-all mandates that could stifle open-source development. Meanwhile, legal scholars emphasize the need for appeal mechanisms and transparency in review decisions to prevent arbitrary enforcement. The debate reflects a broader tension between innovation and oversight that has defined previous tech revolutions—from nuclear energy to biotechnology.

What remains unresolved is how the U.S. will enforce compliance, define “high-risk” systems, and coordinate with international partners. As AI capabilities evolve at breakneck speed, the success of any pre-release regime will depend on its adaptability. The White House is expected to release a formal proposal by late 2024, potentially ahead of a global AI safety summit hosted by South Korea. One thing is clear: the era of unregulated frontier AI may soon be over.

Source: The New York Times