AI Regulation in the European Union and the USA

Definition

The term artificial intelligence has existed for several decades, but its real boom has been recorded in the last five years due to the development of powerful computing capabilities and advances in the field of algorithms. Historically, artificial intelligence began to take shape in the 1950s, when scientists Alan Turing, John McCarthy and Marvin Minsky laid the scientific and technical foundations of artificial intelligence. The term artificial intelligence was first used by John McCarthy in 1956 at a seminar called the “Dartmouth Summer Research Project on Artificial Intelligence.” When examining this concept, we are confronted with different opinions, due to the absence of a universal legal definition and the existence of a large number of ideas, definitions, approaches and theories that are either too inclusive or too specific for a particular sector. Currently, there is no global legal definition of artificial intelligence, and we encounter definitions that are more applicable to regional or local clusters of needs. Several developed countries are currently not interested in international regulation of artificial intelligence, while their relations regarding the development of new technologies are marked by competition (USA versus China) for leadership in this area. At the European Union level, early soft-law instruments from 2018 onwards provided descriptive definitions of artificial intelligence. In the communications “Artificial Intelligence for Europe” (COM(2018) 237 final) and “Coordinated Plan on Artificial Intelligence” (COM(2018) 795 final), artificial intelligence was described as “systems that display intelligent behaviour by analysing their environment and taking action—with some degree of autonomy—to achieve specific goals”. These descriptions served as policy guidance rather than binding statutory definitions. By contrast, the binding Artificial Intelligence Act of 2024 does not contain a single exhaustive legal definition of “artificial intelligence” expressis verbis; instead, it defines the operational concepts “artificial intelligence system” (Article 3(1)), “general-purpose artificial intelligence model” (Article 3(63)) and “general-purpose artificial intelligence system” (Article 3(66)). The distinction between early advisory descriptions and the later hard-law classification of systems is therefore important for understanding the evolution of EU regulatory language. When it comes to AI legislation, the United States has chosen a fragmented, industry-specific approach. There is no single overarching federal AI law. Rather, limited federal statutes, agency guidelines, executive orders, voluntary frameworks, and an increasing number of state-level laws have all contributed to the evolution of policy.

1. Introduction

We encounter the term artificial intelligence in various spheres of our private and working lives, as it affects a wide range of areas in public administration, healthcare, banking, education, industry, transport, etc. It is an opportunity and a challenge for society [1] and at the same time a threat to the violation of fundamental rights and freedoms protected by supranational and national legislation [2].

The concept of artificial intelligence has existed for several decades, historically, artificial intelligence began to take shape in the 1950s, when scientists such as John McCarthy and Marvin Minsky laid the scientific and technical foundations of artificial intelligence. The term “artificial intelligence” was introduced by John McCarthy in 1955. The following year he convened the Dartmouth Summer Research Project, widely regarded as the founding event of the field [3,4,5].

But we have been witnessing its real boom over the last five years as a result of the development of powerful computing capacities and progress in the field of algorithms [6]. As Hamřik [7] (p. 39) states, “this term was not only chosen as the 2023 word of the year by one of the English language dictionary publishers [8], but at the same time it is increasingly resonating in public discourse.”

In examining this concept, we are confronted with various views (Encyclopaedia Britannica [9]; Šufliarsky [10]; Dobrev [11]; Meteňkanyč, Gyurász [12]; Hamřik [7]; Frankish, Keynes, Ramsey [13]), due to the absence of a universal legal definition and the existence of a large number of notions, definitions, approaches and theories [14] that are either too inclusive or too specific to a particular sector [15]. According to Ivor [14], it is necessary to distinguish between artificial intelligence as a scientific discipline and artificial intelligence as a project—the result of scientific-research or implementation activity. Rassi [16] tends to define the term either according to achievable applications or based on scientific principles. As for the requirements that such a definition should meet, Kľučka [17] (p. 554) emphasizes its inclusiveness, precision, complexity, feasibility and stability; the definition should be neither too “narrow”, since it will not be able to absorb the development of new technologies, the legal regulation will over time begin to “lag behind” and may become obsolete; nor too broad, as this could lead to various interpretive problems in the course of its domestic implementation. According to Scherer [18] (p. 359), “the difficulty in defining artificial intelligence lies not in the concept of artificiality but rather in the conceptual ambiguity of intelligence.”

At present, there is no global legal definition of artificial intelligence, and we encounter definitions that are valid rather for regional or local groupings of needs. As Hamřik [7] (p. 54) states, “in view of the global dimension of artificial intelligence technology, international legal regulation would appear rational,” and according to Kohoutová [19] (p. 1285) also “in particular because of the cross-border nature of artificial intelligence.” As Kľučka [17] (p. 557) emphasizes, several developed states are currently not interested in an international regulation of artificial intelligence, and their relations concerning the development of new technologies are marked by competition (the USA versus China) to gain primacy in this field. Another argument is that even if a new legal regulation were adopted, “it cannot be ruled out that it will become obsolete even before the ink on its document has dried. Such regulatory gaps exist because the law is unable to keep pace with technological progress, and its gaps are all the greater the faster technological progress is, with this discrepancy being particularly visible in the field of artificial intelligence.”

The aim of this paper is to provide a structured overview of the regulation of artificial intelligence in the European Union and the United States, to compare their distinctive regulatory approaches across successive phases of development, and to discuss the practical implications of these differences for legal certainty, innovation and cross-border compliance.

2. Overview of AI Regulation in the European Union

2.1. Before 2021: Soft Law and Foundations

An early institutional initiative at European level was the European Parliament’s resolution of 16 February 2017 on Civil Law Rules on Robotics (2015/2103(INL)) [20]. The resolution emphasises the general principle of developing robotic technology with a focus on complementing human capabilities rather than replacing them, supporting research and innovation, the need to develop ethical principles and the creation of a new body, the European Agency for Robotics and Artificial Intelligence. It also addressed concrete domains such as autonomous vehicles and drones, care and medical robots, human enhancement, education, employment, environmental effects and civil liability for harm caused by robots.

At the level of the European Union, legislation in this area began to take clear shape from 2018 (this concerns the regulation of artificial intelligence through so-called “soft law”), when the European Commission published two communications: the communication “Artificial Intelligence for Europe” (COM(2018) 237 final) [21] and the communication “Coordinated Plan on Artificial Intelligence” (COM(2018) 795 final) [22]. In these communications, artificial intelligence is defined as “systems that display intelligent behaviour by analysing their environment and taking action—with some degree of autonomy—to achieve specific goals”. The “Ethics Guidelines for Trustworthy Artificial Intelligence” document was approved by the High-Level Expert Group on Artificial Intelligence in the spring of 2019 [23]. In the context of the creation and application of artificial intelligence, these principles place a strong emphasis on respect for fundamental human rights and freedoms. They state that “AI systems need to be human-centric, resting on a commitment to their use in the service of humanity and the common good, with the goal of improving human welfare and freedom” [23] (p. 4). In parallel with this document, the expert group prepared the document “A Definition of AI: Main capabilities and scientific disciplines” [24], in which it elaborated the definition of this concept proposed in the European Commission’s communication “Artificial Intelligence for Europe” (COM(2018) 237 final). It views artificial intelligence as a system and a scientific discipline [24]. In the communication “Building Trust in Human-Centric Artificial Intelligence” (COM(2019) 168 final) [25], the European Commission welcomed the contribution of the High-Level Expert Group on Artificial Intelligence and stressed that [25] “the ethical dimension of AI is not a luxury feature or an add-on: it needs to be an integral part of AI development. By striving towards human-centric AI based on trust, we safeguard the respect for our core societal values and carve out a distinctive trademark for Europe and its industry as a leader in cutting-edge AI that can be trusted throughout the world.”

In order to create a coordinated framework for the responsible, safe and ethical use of artificial intelligence, on 19 February 2020, the European Commission adopted the document “White Paper on Artificial Intelligence–A European approach to excellence and trust” (COM(2020) 65 final), which acknowledges that artificial intelligence is “a collection of technologies that combine data, algorithms and computing power” [26]; the document presents AI as a strategic technology—understood as a combination of data, algorithms and computing power—whose benefits for citizens, businesses and society depend on a human-centric, ethical and rights-respecting approach [26]. Following the release of the White Paper and the Report on the safety and liability implications of Artificial Intelligence, the Internet of Things, and robotics (COM(2020) 64 final) [27], the European Commission began consulting with Member State stakeholders (civil society, industry, and academia) on particular recommendations for the European approach to AI.

The online consultation ran from 19 February to 14 June 2020 and attracted 1215 responses. Participants generally called for a clear and precise definition of AI and for clarification of key concepts such as risk, high-risk and harm. A clear majority favoured a risk-based approach over uniform regulation of all AI systems. These soft-law instruments responded to growing concerns about the use of AI in high-stakes domains (including biometric identification, automated decision-making affecting fundamental rights, and li-ability for autonomous systems), and sought to prepare a coherent European approach before binding legislation was proposed.

Report on the safety and liability implications of Artificial Intelligence, the Internet of Things and robotics (COM(2020) 64 final) [27] is a companion document to the White Paper on Artificial Intelligence [26]. It also sought to identify and examine the wider implications and possible gaps in liability and safety frameworks for artificial intelligence, the Internet of Things and robotics. This obligation followed from the Communication “Artificial Intelligence for Europe” (COM(2018) 237 final), in which the European Commission committed to assessing the impact of new digital technologies on existing safety and liability frameworks [21].

In October 2020, the European Parliament adopted three resolutions in the field of artificial intelligence concerning ethics [28], liability [29] and intellectual property rights [30]. In 2021, these were followed by resolutions on artificial intelligence in criminal matters [31] and in education, culture and the audiovisual sector [32], making it the first European Union institution to begin addressing the practical application of artificial intelligence by public authorities in the Member States—in particular by police and judicial authorities in criminal matters [10]. The European Economic and Social Committee [33], the European Council [34] and the Council of the European Union [35] also issued opinions and conclusions supporting a human-centric and trustworthy approach to AI. The European Parliament further addressed questions of interpretation and application of international law in the context of artificial intelligence [36]. These soft-law steps and institutional positions prepared the ground for the formal legislative proposal of April 2021 and the updated coordinated plan [37].

2.2. 2021–2025: Legislative Process and Entry into Force of the AI Act

On 21 April 2021, the European Commission presented a proposal for rules and measures [38] aimed at achieving excellence and trust in artificial intelligence, thereby building on many years of work by expert groups and on the demands of European Union institutions in the form of various opinions and recommendations [33,34,35,36]. It took the view that the combination of the proposed legal framework for artificial intelligence and the updated coordinated plan [37] would guarantee safety and the protection of fundamental human rights, as well as legal certainty for businesses in the EU Member States. This proposal represented the official beginning of the legislative process, which was de facto completed in December 2023, when the EU institutions (the European Parliament and the Council of the EU) reached a political agreement on the final wording of the Artificial Intelligence Act [39]. The Artificial Intelligence Act [39] was published in the Official Journal of the European Union on 12 July 2024 (OJ L, 2024/1689, 12 July 2024), came into force on 1 August 2024. According to point 179 of the recital, it applies from 2 August 2026, with staggered application dates [40].

The prohibitions on certain AI practices and the AI literacy obligations started to apply on 2 February 2025; governance rules and the obligations for general-purpose AI models began to apply from 2 August 2025; and the rules applicable to high-risk AI systems that are embedded in regulated products benefit from an extended transition period until 2 August 2028 [40]. These binding transition periods under the AI Act should be clearly distinguished from the Commission’s subsequent proposal for a “Digital Omnibus on AI” (COM(2025) 836 final) [41], which seeks targeted simplifications, particularly for SMEs, and may further adjust certain implementation arrangements. The proposed Omnibus measures remain subject to the ordinary legislative procedure and do not themselves alter the deadlines already laid down in the AI Act.

Despite the absence of a legal definition of the term artificial intelligence expressis verbis, the key concepts are “artificial intelligence system” (Article 3(1)), “general-purpose artificial intelligence model” (Article 3(63)) and “general-purpose artificial intelligence system” (Article 3(66)) [39].

Although the AI Act does not contain a single exhaustive statutory definition of “artificial intelligence” expressis verbis, Article 3(1) defines an “artificial intelligence system” as a “machine-based system designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers from the input it receives how to generate outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments” [39]. This operational definition has been the subject of considerable debate concerning its breadth and the extent to which it captures conventional automated software versus more advanced adaptive systems; Recital 12 of the AI Act distinguishes AI systems from simpler traditional software systems based solely on rules defined by natural persons, and the European Commission has issued dedicated guidelines clarifying that the capacity to infer is the principal distinguishing feature [39,42]. Complementary definitions of “general-purpose artificial intelligence model” (Article 3(63)) and “general-purpose artificial intelligence system” (Article 3(66)) further delineate the regulatory scope of the most powerful models [39].

2.3. Beyond 2025: Implementation, Enforcement and Operational Realities

The period after 2025 is decisive for the practical effectiveness of the AI Act. Prohibited practices and AI literacy obligations have been in force since February 2025, and GPAI model obligations since August 2025. High-risk systems face a longer transition until August 2028. The newly established AI Office within the European Commission, together with national market surveillance authorities, is responsible for supervision and enforcement. Maximum administrative fines can reach €35 million or 7% of worldwide annual turnover for the most serious infringements of prohibited practices, and €15 million or 3% for other obligations, creating significant compliance incentives [39].

Cross-border enforcement and the extraterritorial application of the Act (the so-called Brussels Effect) already influence the design and risk-management practices of non-EU providers that place AI systems on the Union market or whose outputs are used in the Union [43]. Separately from the binding transition periods fixed in the AI Act, the 2025 “Digital Omnibus on AI” proposal [41] indicates that the co-legislators are prepared to introduce targeted simplifications, particularly for SMEs, without abandoning the core risk-based architecture. Future guidance from the AI Office, harmonised standards and the first enforcement decisions will determine whether the framework can deliver both robust protection of fundamental rights and workable conditions for innovation.

3. Overview of AI Regulation in the USA

The United States has adopted a decentralised, sector-specific regulatory strategy to AI regulation. There is no single overarching federal AI law. Instead, policy has evolved through executive orders, voluntary frameworks, limited federal statutes, agency guidance, and a growing body of state-level laws [44].

3.1. Before 2021: Early Coordination and National Strategy

In October 2016, the National Science and Technology Council (NCTS) published the report Preparing for the Future of Artificial Intelligence [45]. The report favoured a risk-based approach guided by public-safety considerations and supported research and development with only limited restrictions. It was prepared by the NSTC Sub-Committee on Machine Learning and Artificial Intelligence, set up earlier that year to improve coordination across federal agencies and to follow AI developments in industry, research and government. The report follows a series of public outreach activities led by The Office of Science and Technology Policy (OSTP), designed to allow government officials thinking about these topics to learn from experts and from the public [45].

Executive Order on Maintaining American Leadership in Artificial Intelligence was issued on 11 February 2019, by Donald J. Trump as the President of the United States [46].

The National Artificial Intelligence Initiative Act of 2020 (Division E of Pub. L. 116-283) establishes a coordinated federal initiative to accelerate AI research, development, demonstration, standards, and workforce development across U.S. federal agencies. It creates the National Artificial Intelligence Initiative Office (NAIIO) within OSTP, requires interagency coordination, directs the National Institute of Standards and Technology (NIST), the National Science Foundation (NSF), Department of Energy (DOE) and others to support AI research, standards, trustworthiness frameworks, and public outreach, and authorizes reporting and assessments to Congress. The Act became effective on 1 January 2021. A crucial point for businesses and developers is that this law does not impose direct penalties or create new liability rules for private companies using AI. Instead, its “enforcement teeth” are primarily felt through federal grant conditions, procurement requirements, and the adoption of voluntary standards that agencies are encouraged to develop [47,48].

3.2. 2021–2025: The Pendulum of Executive Orders and First Targeted Federal Legislation

Between 2021 and 2025 the federal approach to AI was characterised by successive and partly contradictory executive orders that illustrate the “pendulum swing” of administrative policy. In October 2022, the White House Office of Science and Technology Policy released the non-binding Blueprint for an AI Bill of Rights [49]. The document set out five principles intended to guide the design and use of automated systems:

• Safety and effectiveness,
• Protection against algorithmic discrimination,
• Data privacy,
• Notice and explanation,
• The availability of human alternatives, consideration and fallback options.

These principles were presented as guidance especially where automated systems may affect people’s rights, opportunities or access to essential services. The subsequent executive orders and the first targeted federal statute (the TAKE IT DOWN Act) further illustrate how specific concerns—model safety, civil rights, deepfakes and international competitiveness—have shaped the evolving U.S. response.

Executive Order 14110—Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence of 30 October 2023—directed federal agencies to develop safety standards, require testing of advanced models, strengthen civil-rights protections and use federal procurement as a policy lever [50].

After Donald Trump took office, this Executive Order 14110 was revoked, and the President issued a new Executive Order 14 179—Removing Barriers to American Leadership in Artificial Intelligence on 23 January 2025. The new order sought to remove regulatory obstacles regarded as barriers to American AI innovation and to prioritise the maintenance of global leadership in the field, and therefore it is important to develop AI systems that are free from ideological bias or engineered social agendas [51].

Alongside these shifting executive directives, on 28 April 2025, Congress passed the TAKE IT DOWN Act (S.146, signed 19 May 2025, Public Law 119-12), the first major federal AI-targeted law. It criminalizes nonconsensual publication of intimate images, both authentic and computer-generated (including AI-generated deepfakes) and requires platforms to remove them promptly upon notice [52,53]. This statute provides a more durable legal baseline than temporary executive orders.

On 11 December 2025, another important Executive Order 14365 was issued—Ensuring a National Policy Framework for Artificial Intelligence—directing a “minimally burdensome” national approach, establishing an AI Litigation Task Force to challenge conflicting state laws (on grounds like interstate commerce or First Amendment), and evaluating state laws for pre-emption or funding conditions [54].

3.3. Enduring Statutory and Agency Mechanisms

While executive orders attract considerable attention, more durable elements of U.S. AI governance reside in existing federal statutes and the enforcement practices of independent agencies, as well as in state legislation. The Federal Trade Commission (FTC) has applied Section 5 of the FTC Act to prohibit unfair or deceptive practices involving AI, including “AI washing” (misleading claims about AI capabilities), algorithmic discrimination and privacy violations [55,56]. The Department of Justice (DOJ) addresses civil-rights and antitrust concerns arising from AI systems. Sector-specific regimes such as HIPAA (health data), SEC disclosure rules and consumer-protection statutes continue to apply irrespective of changes of presidential administration. At the state level, California’s CCPA/CPRA and a growing body of state AI and deepfake laws (including California’s Transparency in Frontier Artificial Intelligence Act and related measures) create a patchwork of obligations that often prove more stable than federal executive policy [57,58]. These underlying mechanisms provide greater legal continuity than the successive waves of executive orders.

3.4. Beyond 2025: National Framework and Targeted Security Measures

On 20 March 2026 the White House released the National Policy Framework for Artificial Intelligence [59], containing legislative recommendations organised around seven themes, including protection of children, free speech, innovation and establishing a federal policy framework, pre-empting cumbersome state AI laws. While the framework itself does not create binding legal obligations, its content is likely to shape federal legislation on artificial intelligence in the coming period.

Executive Order 14409—Promoting Advanced Artificial Intelligence Innovation and Security (issued on 2 June 2026, by President Donald Trump) shifted emphasis toward cybersecurity and the protection of advanced AI models while still avoiding broad new technology-sector regulation. This shifts the administration’s policy from pure deregulation to targeted protection of national security and critical infrastructure [60,61]. The combination of a proposed national framework, continued agency enforcement and targeted statutes suggests that the United States is moving, albeit gradually, toward a more coherent but still predominantly innovation-oriented and agency-driven model.

4. Discussion

The contrasting regulatory trajectories of the European Union and the United States reflect deeper differences in legal culture, institutional design and political economy. The EU’s preference for comprehensive, rights-protective legislation is rooted in a civil-law tradition that privileges ex-ante rules and a strong constitutional commitment to fundamental rights. The U.S. preference for sector-specific rules, agency enforcement and limited federal legislation is consistent with a common-law tradition, a federal structure that leaves substantial room to the states, and a political culture that places a high premium on technological leadership and private-sector innovation. These structural differences explain why the EU moved relatively quickly from soft-law instruments to a binding horizontal regulation, while the United States has continued to rely on a combination of executive action, existing agency authority and targeted statutes.

When the evolution of each jurisdiction is examined through the three temporal phases (before 2021, 2021–2025, and beyond 2025), the pattern becomes clearer. Both regions began with soft-law and strategic documents. Between 2021 and 2025, the EU completed a major legislative project, whereas the United States experienced rapid policy oscillation through successive executive orders and adopted only one major AI-specific federal statute. After 2025, the EU entered an implementation and enforcement phase, while the United States began to articulate a more coherent national framework while still emphasizing deregulation and national security.

From an industry perspective, the EU AI Act creates higher compliance costs and legal certainty, particularly for high-risk and general-purpose systems, while the U.S. model offers greater flexibility at the price of regulatory fragmentation and uncertainty about future executive policy. Multinational firms must therefore navigate both regimes simultaneously. In this sense, the successive regulatory instruments on each side of the Atlantic are not merely formal lists of measures; they reflect different answers to shared problems (risk of harm, accountability, innovation capacity) and produce different patterns of compliance burden, enforcement style and extraterritorial reach.

Table 1 provides a comparative overview of the main features and implications of the two regulatory approaches:

Table 1. Comparative Overview of EU and US AI Regulatory Approaches.

Aspect	EU AI Act	US Approach	Key Implications
Regulatory philosophy	precautionary; rights-based	innovation-first; market-driven	EU offers higher legal certainty; US enables faster innovation
Structure	horizontal; comprehensive risk classification	decentralised; sector-specific; executive orders	EU provides uniformity; US creates fragmentation
Risk management	ex-ante prohibitions and strict obligations	voluntary frameworks and targeted legislation	EU better prevents harm; US more adaptable
Innovation impact	higher compliance costs, especially for SMEs	lower regulatory burden attracts investment	potential “Brussels Effect” vs. US tech. leadership
Enforcement	high fines (in some cases up to 7% total worldwide annual turnover),	agency-based and state level	strong deterrents in EU; flexibility in US
Adaptability	slower to embrace technological change	high adaptability	EU may lag in frontier models; US leads
Global influence	strong (extraterritorial reach)	moderate (via standards and investment)	shaping of international norms

4.1. The Brussels Effect and Implications for Global Audiences

A central transnational implication of the EU AI Act is the so-called Brussels Effect: the capacity of EU regulation to shape the global behaviour of firms that wish to retain access to the European market [43]. Because many leading AI developers are U.S.-based multinationals, the Act’s extraterritorial reach effectively requires them to adapt model development, risk-management systems and documentation practices to EU standards even for products primarily destined for other markets. In this sense, the EU functions as an indirect global regulator. Policymakers in third countries can therefore benefit from close monitoring of the Act’s implementation, both as a source of technical standards and as a potential template for their own risk-based frameworks.

At the same time, the U.S. experience demonstrates the advantages of flexible, agency-driven oversight and rapid policy adjustment. Jurisdictions that lack the institutional capacity or political consensus for a comprehensive horizontal statute may prefer to strengthen existing consumer-protection, data-protection and sectoral regulators while developing targeted rules for high-risk applications. Hybrid approaches that combine elements of risk classification with sector-specific enforcement are likely to prove attractive to many emerging and mid-sized economies.

4.2. Suggestions for Future Research

The comparative analysis of the EU and U.S. approaches across the three temporal phases opens several promising avenues for future research. In particular, the present Entry provides a structured overview of regulatory instruments and approaches; deeper empirical and socio-legal work is needed to examine the regulatory incidents that prompted intervention, the interactions between regulators and regulated actors, and the differentiated effects of AI regulation across sectors and technologies.

• Empirical assessment of regulatory outcomes. Longitudinal studies should measure the actual effects of the AI Act’s phased implementation (2025–2028) on innovation rates, compliance costs, market entry of SMEs, and the incidence of AI-related harms in the European Union and compare these outcomes with the effects of successive U.S. executive orders, the TAKE IT DOWN Act, and state-level legislation on investment, product development and risk mitigation.
• Regulatory incidents and triggers of intervention. Future research should systematically identify the concrete regulatory incidents—such as high-profile failures of biometric systems, harmful deepfakes, discriminatory automated decision-making, or deceptive “AI washing”—that have shaped legislative and enforcement responses in the EU and the United States and assess which types of incidents have proven regulation-relevant and which have not.
• Interactions between regulator, regulated actors, incident and outcome. Socio-legal and process-tracing studies could interrogate the complex interactions between regulators, regulated firms, the incidents that prompted action, and the resulting regulatory outcomes, thereby moving beyond catalogues of instruments toward an understanding of how regulation of AI actually unfolds in practice.
• Sector- and technology-specific regulatory patterns. Treating AI as a combination of technologies rather than a single technique, comparative research should examine how regulatory interventions differ across sectors (e.g., healthcare, finance, employment, law enforcement, content platforms) and across classes of systems (narrow automated tools, high-risk applications, general-purpose models), and what implications these differences have for industry and users.
• Comparative enforcement research. Systematic comparison of how the European AI Office and national market surveillance authorities on the one hand, and the FTC, DOJ and state attorneys general on the other, handle analogous cases involving algorithmic discrimination, deepfakes, general-purpose AI models and “AI washing” would clarify the practical strengths and weaknesses of ex-ante versus ex-post enforcement models.
• The Brussels Effect in practice. Quantitative and qualitative research is needed to determine the extent to which the EU AI Act influences model design, risk-management systems and contractual practices of non-EU (particularly U.S.) providers, and whether this extraterritorial influence leads to de facto global standard-setting or to regulatory fragmentation and dual-compliance strategies.
• Hybrid regulatory models. Investigation of hybrid frameworks that combine elements of the EU’s risk-based classification with the U.S. reliance on existing agency authority and targeted statutes could identify designs better suited to mid-sized and emerging jurisdictions that seek both rights protection and innovation incentives.
• Public trust and legitimacy. Longitudinal surveys and experimental studies examining how different regulatory regimes affect public trust in AI systems, perceived legitimacy of oversight institutions, and support for stricter or more flexible rules would help evaluate the social sustainability of each approach.
• Interaction with adjacent legal regimes. Further research should examine how AI-specific rules interact with data-protection law (GDPR and state privacy statutes), competition law, intellectual-property regimes and national-security frameworks in both jurisdictions, and whether these interactions create synergies or regulatory conflicts.
• Geopolitical and competitive dynamics. Analysis of how the EU–U.S. regulatory divergence (and possible future convergence) affects global AI competition, particularly in relation to China, and whether regulatory choices become instruments of technological and geopolitical strategy.

5. Conclusions

This entry paper has examined the development of AI regulation in the European Union and the United States across three temporal phases. The analysis highlights two distinct paradigms: the EU’s horizontal, risk-based AI Act and the U.S.’s decentralised, innovation-driven framework that rests on a combination of executive action, agency enforcement and targeted statutes.

Neither model is inherently superior; each embodies a fundamental trade-off between preventive governance and technological dynamism. The EU offers greater legal certainty and protection of rights at the potential cost of speed of innovation, while the United States encourages rapid development but faces the challenges of fragmentation and policy volatility. The Brussels Effect means that the practical reach of the EU AI Act already extends well beyond Europe’s borders [43].

Looking ahead, greater convergence may emerge around the regulation of frontier models and high-stakes applications such as deepfakes. Policymakers in other regions can benefit from both experiences by combining clear risk classification with flexible, innovation-friendly instruments.

Policy recommendations include simplifying compliance pathways for SMEs within the EU framework, enhancing federal coordination in the United States while preserving innovation incentives, and strengthening transatlantic cooperation on shared priorities such as AI safety and ethical standards.

Ultimately, the regulation of artificial intelligence will play a decisive role in determining not only the pace of technological progress but also the values embedded in future AI systems. A balanced approach that safeguards human rights without stifling innovation remains the central challenge for policymakers on both sides of the Atlantic and beyond.