How Should OpenAI Be Regulated?

Last updated 3 months ago. Our resources are updated regularly but please keep in mind that links, programs, policies, and contact information do change.

The November 2022 release of ChatGPT thrust artificial intelligence from science fiction into daily reality for millions of Americans. The conversational AI tool, developed by OpenAI, sparked a global debate about one of the most consequential technologies of our time.

OpenAI sits at the center of this transformation. The company has evolved from an idealistic nonprofit research lab into a global tech powerhouse worth billions of dollars. Its partnership with Microsoft and rapid-fire product releases have made it the face of the AI revolution.

But OpenAI’s meteoric rise has triggered intense scrutiny from government regulators. The technology presents both immense potential benefits and unprecedented risks. Policymakers are grappling with fundamental questions: Should the government regulate advanced AI systems? How can America maintain its technological edge while preventing catastrophic harm?

OpenAI: From Nonprofit to Tech Giant

OpenAI’s unique corporate structure and mission are central to the regulatory debate. The company’s evolution from idealistic nonprofit to commercial powerhouse illustrates the tension between humanitarian goals and business realities.

Founding Mission

OpenAI launched in December 2015 as a nonprofit artificial intelligence research laboratory. Tech visionaries including Sam Altman, Greg Brockman, Ilya Sutskever, and Elon Musk founded the organization with an ambitious mission: to ensure that artificial general intelligence (AGI) “benefits all of humanity”.

The organization’s charter explicitly acknowledged that AGI – defined as “highly autonomous systems that outperform humans at most economically valuable work” – could damage society if “built or used incorrectly.” OpenAI positioned itself as a research institution dedicated to steering AI development toward positive outcomes for everyone, not just shareholders or a single entity.

This idealistic, safety-first mission soon collided with practical realities. Developing powerful AI models requires enormous resources, particularly computational power and top-tier research talent. By 2019, it became clear that a purely nonprofit model relying on donations could not compete with the deep pockets of corporate giants like Google and Meta.

Corporate Transformation

This led to a pivotal strategic shift in 2019. OpenAI transitioned from a nonprofit to a “capped-profit” entity. This hybrid model created a new for-profit arm, OpenAI LP (later OpenAI Global, LLC), which could attract venture capital investment and offer employees equity stakes.

To preserve its original mission, the company “capped” profits for investors – initially at 100 times their investment – with any excess returns flowing back to the original nonprofit parent. This structure was designed to legally bind the for-profit arm to the nonprofit’s charter, requiring all investors and employees to prioritize the mission of safe and beneficial AGI over financial interests.

The nonprofit OpenAI, Inc. remains the sole controlling shareholder of its for-profit subsidiaries. A majority of the nonprofit’s board members are barred from holding financial stakes in the for-profit arm, a rule intended to ensure their fiduciary duty remains aligned with the original charter.

Microsoft Partnership

This structure enabled the most critical partnership in modern AI: OpenAI’s alliance with Microsoft. Beginning with a $1 billion investment in 2019 and growing to a total of $13 billion, Microsoft became OpenAI’s essential partner.

The partnership is symbiotic. Microsoft provides vast cloud computing resources via its Azure platform, which are indispensable for training and deploying large-scale AI models. In return, Microsoft received a 49% stake in OpenAI Global’s profits (subject to the cap), an exclusive license to integrate OpenAI’s technologies into products like Bing and Microsoft 365, and a non-voting observer seat on the board.

While this deep integration has been crucial for both companies, OpenAI’s governance structure formally maintains that Microsoft does not have control over its operations or decision-making processes.

Revolutionary Products

OpenAI’s research breakthroughs have translated into pioneering products that redefined AI’s possibilities. The company focuses on generative AI – models capable of creating new content rather than simply analyzing existing data.

The most famous products are the Generative Pre-trained Transformer (GPT) series of large language models. These models, including GPT-3 and GPT-4, power tools like ChatGPT. The release of ChatGPT demonstrated unprecedented ability to understand and generate human-like text, answering questions, writing essays, and composing code.

Beyond text, OpenAI has pushed into other domains. The DALL-E series generates complex and creative images from simple text descriptions, while the Sora model can create high-fidelity video clips from text prompts, showcasing rapid advancement in generative AI capabilities.

The rapid acceleration shown in these milestones helps explain the urgency felt by policymakers. The technology has moved from niche research concept to globally impactful product in just a few years, forcing a reactive scramble to understand and manage its consequences.

The Regulatory Debate

The debate over regulating OpenAI and the broader AI industry represents a clash of competing worldviews about risk, innovation, and government’s proper role. Both sides invoke national interest and societal well-being, but they fundamentally disagree on where the greatest danger lies.

The Case for Strong Regulation

The most urgent calls for government intervention are rooted in belief that advanced AI’s potential harms are too great to be left to corporate self-regulation. These arguments span from immediate societal problems to long-term existential threats.

Catastrophic and Existential Risk: Proponents of strong regulation, including OpenAI’s own leaders, have publicly warned that future “superintelligence” – an AI far surpassing human intellect – could pose risks to humanity’s survival. In a notable blog post, CEO Sam Altman and co-founders called for creating an international regulatory body for AI, similar to the International Atomic Energy Agency. Such an agency would inspect advanced AI systems, require audits, test for compliance with safety standards, and restrict the most powerful models to prevent accidental or deliberate creation of technology with power to destroy its creators.

Ethical Use and Human Rights: Advocates argue that profit motives will inevitably lead companies to cut corners on safety and ethics unless compelled by law. This includes several key areas:

Preventing Bias and Discrimination: AI models trained on vast internet datasets can absorb and amplify existing societal biases related to race, gender, and other characteristics. Regulation is seen as necessary to mandate that companies actively identify and remove these biases to avoid reproducing discrimination in critical areas like hiring, lending, and criminal justice.

Protecting Privacy: AI systems pose significant privacy risks. Proponents argue for need to adapt and rigorously enforce existing digital privacy laws to ensure personal information isn’t collected, used, or exposed harmfully by AI models.

Combating Misinformation and Malicious Use: Generative AI makes it easy to create hyper-realistic “deepfakes,” spread disinformation, and conduct sophisticated phishing attacks. Critics contend that tech companies have been unwilling, not unable, to effectively police this misuse. Government intervention is required to create and enforce rules countering erosion of public trust and protecting democratic processes from manipulation.

Enforcing Existing Laws: A more pragmatic argument is that AI provides new, powerful tools for committing old abuses at unprecedented scale. Federal law already prohibits fraud, market manipulation, and discrimination in housing and employment. Regulation is needed to clarify how these existing laws apply to AI-driven systems and ensure companies cannot use AI’s complexity as a shield to evade responsibility for illegal conduct amplified by their technology.

The Case for Light-Touch Regulation

Opposing stringent oversight is a powerful coalition arguing that heavy-handed regulation poses a greater threat than the technology itself. Their arguments are grounded in principles of economic competitiveness, free-market innovation, and individual liberty.

Innovation and Competitiveness: The dominant argument against regulation is that it will stifle innovation and harm U.S. competitiveness. This perspective frames AI development as a geopolitical “race,” primarily against China. Proponents warn that bogging down American developers in complex bureaucratic red tape and compliance costs will hand decisive advantage to state-backed competitors in China, who operate with fewer restrictions. Maintaining America’s technological leadership is viewed as a national security imperative that outweighs hypothetical technology risks.

Avoiding “Regulatory Capture”: Some critics, particularly from venture capital and startup communities, argue that the loudest calls for regulation often come from the largest, most entrenched corporations. They contend that these giants can easily absorb high compliance costs, while smaller startups cannot. Complex regulation would not create a safer AI ecosystem, but a less competitive one. It would create high barriers to entry, crush entrepreneurial spirit, and cement market dominance of companies like Google, Microsoft, and OpenAI itself, ultimately leading to slower product improvement and fewer breakthroughs.

Free Expression Concerns: A third argument, rooted in constitutional principles, views AI regulation as a threat to free expression and the marketplace of ideas. Many proposed regulations – particularly those aimed at controlling “harmful” content like misinformation – are seen as a new front in battles over content moderation and censorship. Proponents argue that AI is a powerful tool for communication and creativity. Government mandates on what AI models can say, or requirements to label AI-generated content, could represent dangerous government intrusion into speech protected by the First Amendment. The proper remedy for false or harmful speech is not government censorship but more speech and an educated, discerning public.

OpenAI’s Strategic Position

OpenAI’s stance on regulation is complex and seemingly contradictory. The company has publicly positioned itself as a leading advocate for government oversight while simultaneously lobbying for policies that would significantly limit the scope and power of that oversight.

Public Call for Global Watchdog: OpenAI has been notably proactive in its public calls for regulation. In high-profile Congressional appearances and published essays, CEO Sam Altman has argued for creating an international agency to govern “superintelligent” AI. This proposed body would be granted significant powers, including authority to inspect advanced AI systems, conduct audits, test for compliance with safety standards, and restrict any AI model surpassing certain capability thresholds.

By publicly framing risks in stark, existential terms, OpenAI accomplishes several strategic goals. It positions the company as a responsible steward of powerful and potentially dangerous technology. It demonstrates willingness to partner with governments to find solutions. It focuses regulatory conversation on a long-term, highly complex international challenge – governance of hypothetical future superintelligence. This grand vision, while important, is unlikely to result in immediate, binding regulations on OpenAI’s current products.

Private Lobbying for “Freedom to Innovate”: While publicly discussing need to regulate future superintelligence, OpenAI’s direct policy advocacy has focused on securing a more permissive environment for its current operations. The company’s formal proposals reveal a clear agenda centered on “freedom to innovate.”

A central plank is the call for federal preemption of state laws. OpenAI has directly urged the White House and Congress to pass federal legislation that would supersede the burgeoning “patchwork” of state-level AI regulations. The company argues that complying with potentially 50 different sets of rules would be unmanageable and would “bog down innovation.” In a formal submission to the White House, OpenAI proposed that in exchange for voluntary cooperation with federal bodies like the U.S. AI Safety Institute, private companies should receive “relief from the 781 and counting proposed AI-related bills already introduced this year in US states.”

Another critical lobbying point concerns copyright law. OpenAI has advocated for an expansive interpretation of the “fair use” doctrine, arguing that its AI models must be allowed to train on vast amounts of copyrighted material to remain competitive. The company frames this not as commercial convenience but as national security imperative, warning that if U.S. copyright laws are too restrictive, American AI development will be crippled, ceding the field to Chinese competitors who face no such constraints.

Government Response: Executive Branch

The executive branch has responded to the rise of generative AI with a decisive policy pivot, moving from promoting AI in a technology-neutral manner to pursuing an aggressive strategy aimed at accelerating development and ensuring American dominance.

America’s AI Action Plan

In July 2025, the White House unveiled “Winning the AI Race: America’s AI Action Plan,” a sweeping policy roadmap fulfilling President Trump’s January 2025 executive order on “Removing Barriers to American Leadership in Artificial Intelligence.” The plan’s stated purpose is unambiguous: to “cement U.S. dominance in artificial intelligence” by aggressively cutting red tape, removing regulatory burdens, and supercharging public and private investment in AI infrastructure and talent.

The plan is organized around three central pillars:

Accelerating Innovation: Fostering rapid development and deployment of AI technologies across the economy.

Building American AI Infrastructure: Strengthening foundational components – from data centers to the energy grid – necessary for AI advancement.

Leading in International Diplomacy and Security: Shaping global AI norms and protecting U.S. national security interests in the AI landscape.

Within these pillars, the plan outlines over 90 specific policy actions:

Widespread Deregulation: The plan directs federal agencies to conduct thorough review of existing regulations and identify and repeal any rules perceived as hindering AI development and deployment. It also uses federal funding as leverage, encouraging states to adopt similar deregulatory posture by making federal AI-related grants and resources contingent on their refraining from imposing new regulatory requirements.

Rapid Infrastructure Buildout: To address the immense physical needs of the AI industry, the plan calls for expediting and modernizing permitting processes for essential infrastructure, particularly data centers and semiconductor fabrication plants. This includes efforts to streamline environmental reviews that have historically slowed such projects. It also calls for modernizing the national power grid to meet rising energy demands of advanced AI systems.

Exporting American AI: The plan establishes a proactive strategy to export “full-stack” American AI packages – including hardware, models, software, and standards – to allied nations. The goal is to expand U.S. influence and entrench American technology and values in the global AI ecosystem, while simultaneously tightening export controls on advanced AI technology flowing to strategic competitors like China.

Promotion of Open-Source AI: The plan expresses strong policy preference for open-source and open-weight AI models, whose underlying code and data are publicly available. It encourages federal agencies to prioritize use and development of such models, a shift from the more cautious approach of the previous administration.

Key Executive Orders

President Trump signed three executive orders that put the Action Plan’s policies into immediate effect and introduced a new, distinctly ideological dimension to U.S. AI policy.

The most notable is the order on “Preventing Woke AI in the Federal Government.” This marks the first time the U.S. government has explicitly attempted to shape the ideological behavior of AI models. The order directs that any AI system procured by the federal government must be “objective and free from top-down ideological bias.” It specifically targets what it calls the “destructive” ideology of diversity, equity, and inclusion (DEI), and lists concepts like “critical race theory, transgenderism, unconscious bias, intersectionality, and systemic racism” as examples of prohibited biases.

This represents a sharp reversal from the previous administration’s focus on using AI policy to protect against algorithmic discrimination and ensure equitable outcomes. This new policy forces tech companies seeking lucrative government contracts to navigate a politically charged culture war, creating pressure to self-censor their models to align with the administration’s definition of ideological neutrality.

A second order on “Accelerating Federal Permitting of Data Center Infrastructure” gives teeth to the Action Plan’s infrastructure goals. It specifically targets long-standing environmental laws, such as the National Environmental Policy Act (NEPA), aiming to streamline the review process for data centers by granting them “categorical exclusions” from certain requirements. While framed as necessary to maintain America’s competitive edge, this move has drawn sharp criticism from environmental groups, who argue it weakens foundational law designed to protect communities and the environment from impacts of large-scale industrial projects.

Government Response: Legislative Branch

Congress presents a more complex picture than the executive branch’s clear strategy. Lawmakers are grappling with the same fundamental tensions as the rest of the country, resulting in focus on consensus-building “enabling” legislation rather than comprehensive, restrictive regulation.

State vs. Federal Authority

The question of who should regulate AI – the federal government or states – has become a central point of conflict. AI companies, led by OpenAI, have lobbied intensely for federal preemption, arguing that a single, unified federal framework is essential for innovation. They contend that navigating a “difficult to imagine” compliance nightmare of 50 different sets of state regulations would be prohibitively costly and complex, especially for startups, and would ultimately slow the entire U.S. AI sector.

In the absence of comprehensive federal action, states have embraced their traditional role as “laboratories of democracy.” Since 2019, state legislatures have become hotbeds of AI policy experimentation. All 50 states, along with several territories, have introduced AI-related legislation in the 2025 session, with over a hundred laws already enacted in recent years and more than 1,000 bills currently under consideration.

These state laws often target specific, tangible harms that resonate with local constituencies, such as prohibiting deceptive deepfakes in political campaigns, requiring transparency in automated decision-making by state agencies, or regulating AI use in healthcare and insurance. Proponents argue this state-led approach allows for more nimble and responsive policymaking tailored to local needs and values, without waiting for a gridlocked Congress to act.

The tension between these approaches came to a head with the near-enactment of a federal provision that would have imposed a ten-year moratorium on states’ ability to regulate AI. The measure, originally passed by the House of Representatives, ultimately failed in the Senate after facing bipartisan opposition from lawmakers, governors, and conservative groups who viewed it as federal overreach. The collapse was a significant victory for states’ rights advocates and a major setback for the tech industry’s push for preemption.

Current Legislative Proposals

An examination of key AI-related bills moving through the 119th Congress reveals a clear pattern: lawmakers are focused on legislation that enables and supports the AI ecosystem rather than imposing hard limits or restrictions. The most prominent bills enjoy bipartisan support and aim to promote research, foster controlled innovation, and address workforce adaptation.

The CREATE AI Act of 2025 (H.R. 2385): The “Creating Resources for Every American To Experiment with Artificial Intelligence Act” is a bipartisan bill with a primary goal of democratizing access to AI development tools. It finds that immense computational resources and large datasets required for cutting-edge AI research are currently concentrated in the hands of a few large technology companies, limiting research community diversity. To address this, the bill would establish a National Artificial Intelligence Research Resource (NAIRR), providing researchers and students from academia, nonprofits, and small businesses with access to these critical resources.

The Unleashing AI Innovation in Financial Services Act (H.R. 4801 / S.B. 2520): This bipartisan bill seeks to balance fostering innovation with protecting consumers in the sensitive financial services sector. It would authorize federal financial regulatory agencies to create “regulatory sandboxes” – controlled environments allowing companies to test new AI-driven financial products and services for limited time under close regulatory supervision, but without immediate threat of enforcement actions. The goal is to allow experimentation and learning on both sides, enabling new technology development while giving regulators insight needed to craft informed, effective rules.

The Artificial Intelligence and Critical Technology Workforce Framework Act of 2025 (S.1290): Introduced by Senator Gary C. Peters, this bill reflects growing congressional focus on AI’s human impact. Its title indicates a focus on creating a national framework to address workforce challenges and opportunities presented by AI and other critical technologies. This likely includes initiatives related to skills training, education, and adapting the American workforce for future jobs.

This legislative activity shows that while public debate is often framed around “stopping bad AI,” the political reality in Congress is consensus around “promoting good AI” and managing its societal side effects. The more contentious questions about hard limits, liability frameworks, and content regulation are largely being avoided in favor of consensus-building, enabling measures.

Core Regulatory Battlegrounds

The debate over AI regulation spans multiple legal and policy domains. The core challenge in each area is that AI doesn’t necessarily create entirely new types of problems; rather, it radically scales and complicates existing ones.

National Security: Dual-Use Technology

Artificial intelligence is a classic dual-use technology: the same capabilities that can strengthen national defense can also be weaponized by adversaries to undermine it. This creates a complex security dilemma for policymakers.

Threat Landscape: Security experts and government agencies have identified several key areas of concern:

AI-Accelerated Cyberattacks: Malicious actors can use generative AI to create more sophisticated and highly tailored phishing emails, write novel malware, and identify software vulnerabilities at speed and scale that could overwhelm traditional cyber defenses.

Information Ecosystem Erosion: The ability to generate hyper-realistic “deepfake” videos and audio, combined with power to create and disseminate disinformation on massive scale, poses direct threat to democratic processes, social cohesion, and public confidence in institutions.

Weapons Development: Experts warn that AI could be used to assemble knowledge and instructions for creating chemical, biological, radiological, or nuclear weapons, accelerating a trend of proliferating threats from non-state actors.

Government Response: The AI Action Plan acknowledges these risks. The plan calls for close coordination between government agencies and frontier AI developers to assess emerging threats, for the Cybersecurity and Infrastructure Security Agency (CISA) to update its incident response playbooks for AI-specific risks, and for measures to secure critical infrastructure from AI-driven attacks or adversarial interference. The Department of Defense is also actively refining its “Responsible AI” and “Generative AI” frameworks to guide ethical development and deployment of AI in military contexts.

However, the dual-use nature means that every advance in using AI for defensive purposes, such as identifying cyber threats, may inadvertently contribute to knowledge for offensive uses.

Economic Impact: Workforce Transformation

The economic impact of generative AI is a paradoxical story of disruption and opportunity. The technology is simultaneously eliminating some jobs, creating new ones, and fundamentally reshaping work across nearly every economic sector.

Job Displacement: There is clear evidence of job displacement. The tech industry has seen significant layoffs, with up to 80,000 roles eliminated as companies integrate AI into core functions like software engineering and IT support. Roles involving repetitive, information-based tasks are particularly vulnerable to automation. Studies have identified occupations like customer service representatives, bookkeepers, translators, and some administrative roles as being at high risk. Long-term projections from Goldman Sachs suggest that AI could eventually replace the equivalent of 300 million full-time jobs globally.

Job Creation and Enhancement: The data also shows that AI is a powerful engine for job creation and wage growth. A 2024 analysis found that over half of all jobs requiring AI skills were outside the traditional tech sector, in fields like marketing, finance, human resources, and even arts. Job postings in these non-tech sectors that list AI skills command a significant salary premium, averaging 28% – or nearly $18,000 more per year – than comparable roles without those requirements.

Augmentation vs. Automation: For the vast majority of professions, AI is currently acting as a powerful tool that augments human workers, automating routine tasks and freeing them to focus on more complex, creative, and strategic work. A study of the AI model Claude found that while it could automate or augment about 25% of tasks across all jobs, augmentation was far more common than full automation for nearly every occupation. This indicates that, at least in the short term, the primary effect is to change how jobs are done rather than eliminate them entirely.

First Amendment: Regulating Code and Content

Any government attempt to regulate the output of generative AI systems immediately runs into the formidable barrier of the First Amendment. The legal consensus is that AI-generated content, as a product of human creativity and expression using a technological tool, is a form of speech protected by the Constitution. Furthermore, courts have long held that computer code itself is a form of speech, adding another layer of protection.

Constitutional Challenges: This has profound implications for commonly proposed AI regulations. Laws that would mandate watermarks or disclaimers on all AI-generated content face serious constitutional challenge as compelled speech. The government generally cannot force a speaker to include a message they don’t wish to convey unless the regulation is narrowly tailored to serve a compelling government interest. Legal experts argue that broad, blanket disclosure requirements for AI content are unlikely to meet this high standard, as they would burden vast amounts of non-malicious, creative, and political speech.

Deepfakes and False Speech: The regulation of “deepfakes” and false political speech is even more fraught. While the potential for AI-generated fakes to deceive voters is a major concern, the Supreme Court has established very strong, though not absolute, protection for false speech, particularly in the political arena. The prevailing legal doctrine, established in cases like United States v. Alvarez, is that the proper remedy for false speech is counterspeech – more speech from citizens, journalists, and campaigns to debunk lies – not government censorship.

Traditional Exceptions Apply: This doesn’t mean all AI-generated content is immune from regulation. Existing traditional exceptions to the First Amendment apply with full force. Speech constituting defamation, fraud, incitement to imminent lawless action, or true threats is illegal regardless of the tool used to create it. The legal principles remain the same; the challenge for law enforcement and courts is one of scale, speed, and attribution in this new technological environment.

Liability: Who Pays When AI Fails?

When an AI system makes a mistake that causes harm – whether it’s a self-driving car causing an accident, a medical diagnostic tool giving wrong diagnosis, or a financial algorithm making ruinous trade – a critical question arises: who is legally responsible?

Traditional Tort Law Challenges: The unique characteristics of AI – its complexity, “black box” opacity, autonomy, and capacity for continuous learning – pose fundamental challenges to traditional tort law. Under standard negligence claims, a victim must prove that a defendant had a duty of care, breached that duty, and that this breach caused harm. With AI, this can be nearly impossible. A victim may not be able to determine why the AI failed, making it difficult to prove breach of duty. Even if failure is identified, the complex supply chain – involving data providers, model developers, software integrators, and end-users – makes it incredibly difficult to pinpoint fault and establish clear causation.

Products Liability Framework: Because of these challenges, many legal experts argue for applying a products liability framework to AI. This legal theory, which evolved over the 20th century to deal with harms caused by new mass-produced technologies like automobiles, shifts focus from producer behavior (negligence) to product safety itself. Under strict products liability regime, a manufacturer can be held liable for harm caused by a “defective” product, regardless of whether they were negligent. This approach would treat AI systems as products and hold developers liable if design defects, manufacturing defects, or failure to warn of non-obvious risks result in harm.

EU Model: A potential model is the European Union’s proposed AI Liability Directive. This directive introduces several novel legal tools to help victims of AI-related harm. First, it creates a rebuttable presumption of causality, which helps ease the victim’s burden of proof. If a claimant can show that a provider failed to comply with safety rules and that this failure was likely linked to harm, the court can presume a causal link, which the defendant must then disprove. Second, it gives courts power to order disclosure of evidence, forcing companies to provide relevant information and logs about high-risk AI systems suspected of causing damage.

Antitrust: New Gatekeepers

The rise of generative AI has triggered significant antitrust concerns among regulators. The core issue is not traditional market monopoly in a single product, but rather potential for a few dominant companies to leverage their control over essential inputs of the AI ecosystem to stifle competition and entrench their power.

Critical Resource Concentration: Generative AI development depends on three critical, highly concentrated resources:

Massive Datasets: Training foundation models requires access to petabytes of data. Large, incumbent tech companies that have spent decades accumulating vast proprietary datasets from their users have significant, perhaps insurmountable, advantage over new entrants.

Specialized Talent: The pool of world-class AI researchers and engineers is small and highly sought after. Dominant firms can use vast resources to hire the best talent, potentially “locking in” expertise needed for major breakthroughs and preventing it from flowing to smaller rivals.

Immense Computational Power: Training and running large AI models requires access to thousands of specialized semiconductor chips (GPUs) and massive data centers. This “compute” is extraordinarily expensive and primarily supplied by a handful of firms. The market for high-end AI chips is dominated by Nvidia, and cloud computing services providing access to this hardware are controlled by Amazon, Microsoft, and Google.

Antitrust Investigations: This concentration has led to antitrust investigations focused on the Microsoft-OpenAI-Nvidia nexus. The U.S. Department of Justice has launched an investigation into Nvidia’s dominance in the AI chip market, examining whether its software and distribution practices unfairly lock customers into its ecosystem. Simultaneously, the Federal Trade Commission is probing the close partnership between Microsoft and OpenAI. The FTC’s concern is that this partnership could be anticompetitive, allowing Microsoft to leverage its control over cloud computing to give OpenAI unfair advantage, and in turn, use its exclusive access to OpenAI’s leading models to entrench its position in other markets.

Regulatory Models and Global Competition

As nations grapple with AI governance, two distinct philosophical approaches are emerging, creating global divergence in regulatory strategy. The United States champions a “soft law” approach centered on voluntary, industry-led risk management. The European Union pioneers a “hard law” approach based on the precautionary principle and legally binding, risk-based rules.

U.S. Approach: NIST Framework

The centerpiece of the current U.S. approach is the AI Risk Management Framework (AI RMF), developed by the National Institute of Standards and Technology (NIST). Released in January 2023, the AI RMF is a voluntary framework, not binding law. It provides organizations with a flexible, structured guide for managing AI risks throughout their entire lifecycle, from design and development to deployment and evaluation.

The framework is built around four core functions:

Govern: Creating a culture of risk management within an organization. This involves establishing clear policies, processes, and lines of accountability to ensure AI risks are consistently and effectively managed in line with legal requirements, ethical principles, and organizational values.

Map: Identifying the context in which an AI system will operate and mapping potential risks associated with that context. This is a process of proactive risk identification that considers the system’s potential impacts on individuals, groups, and society.

Measure: Focusing on the technical side of risk management. This involves applying quantitative and qualitative metrics and assessment tools to analyze, track, and evaluate AI system performance against key characteristics of trustworthiness.

Manage: Taking action. Based on risks identified in the Map function and assessed in the Measure function, this step involves prioritizing those risks and allocating resources to treat, mitigate, or accept them.

The goal is to build “trustworthy AI” with seven key characteristics: valid and reliable; safe; secure and resilient; accountable and transparent; explainable and interpretable; privacy-enhanced; and fair, with harmful biases managed. The framework adopts a “socio-technical” perspective, acknowledging that AI risks are not purely technical but deeply intertwined with societal dynamics and human behavior.

The power of the NIST AI RMF comes not from legal penalties but from its potential to become a de facto industry standard. It’s intended to be integrated into organizational practices and government procurement requirements, creating market-based incentive for adoption.

Global Regulatory Spectrum

The voluntary U.S. model is just one of several regulatory approaches being implemented globally. These models exist on a spectrum from less interventionist to more coercive:

Principles-Based Approach: The lightest-touch approach, where governments or international bodies issue high-level ethical principles (e.g., fairness, transparency, accountability) to guide AI development without prescribing specific technical or legal requirements.

Standards-Based Approach: This model delegates some regulatory authority to independent technical standards organizations. Compliance with standards developed by these bodies can then be used to demonstrate compliance with broader legal requirements.

Risk-Based Approach (The EU Model): The leading example of “hard law” framework. The EU’s AI Act categorizes AI systems into four tiers of risk: unacceptable risk (banned outright, e.g., social scoring), high risk (subject to strict requirements, e.g., AI in medical devices or hiring), limited risk (subject to transparency obligations, e.g., chatbots), and minimal risk (unregulated). This approach is mandatory and backed by significant financial penalties for non-compliance.

Agile/Experimentalist Approach: This approach uses tools like “regulatory sandboxes” to allow companies to test innovative AI products in controlled, real-world environments with regulatory supervision but relaxed enforcement. This model aims to foster innovation while allowing regulators to learn about new technologies before crafting permanent rules.

This global divergence sets up geopolitical competition between regulatory philosophies. Proponents of the U.S. model argue that the EU’s prescriptive “hard law” approach will stifle innovation and cede technological leadership. Conversely, proponents of the EU model argue that the U.S. “soft law” approach amounts to ineffective self-regulation that fails to adequately protect citizens from harms of powerful and opaque technology.

Our articles make government information more accessible. Please consult a qualified professional for financial, legal, or health advice specific to your circumstances.