The U.S. and the EU regulate AI differently

The European Union's AI Act, spanning 113 articles and roughly 50,000 words, is the first comprehensive AI legislation globally, while the U.S. has no federal AI law, relying on executive orders, agency guidance, and state laws, with U.S. states passing 131 AI-related laws in 2024, more than double the 49 enacted in 2023. The EU AI Act bans eight categories of AI practices, including social scoring and real-time facial recognition in public spaces for law enforcement, with exceptions for four categories, whereas the U.S. has banned no AI technology categories. The EU's Article 4 permits commercial AI training on data only where rights holders haven't actively blocked it, contrasting with the U.S. Fair Use doctrine.

Spanning 113 articles https://www.whitecase.com/insight-alert/long-awaited-eu-ai-act-becomes-law-after-publication-eus-official-journal and roughly 50,000 words, the EU AI https://www.fastcompany.com/section/artificial-intelligence Act is the first of its kind globally to provide comprehensive legislation on AI. Meanwhile, the U.S. has not yet passed any federal laws on the matter, relying instead on executive orders https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/ , guidance from agencies https://www.nist.gov/itl/ai-risk-management-framework like the National Institute of Standards and Technology and the Federal Trade Commission https://www.ftc.gov/ai , and a growing collection of state-level bills https://ai-law-center.orrick.com/us-ai-law-tracker-see-all-states/ . In fact, in 2024, U.S. states passed 131 AI-related laws https://hai.stanford.edu/assets/files/hai ai-index-report-2025 chapter6 final.pdf , more than double the 49 that were enacted in 2023. At first glance, the U.S. appears more active, but when you measure the weight of its AI laws, the picture reverses. For instance, a California deepfake bill https://www.gov.ca.gov/2024/09/17/governor-newsom-signs-bills-to-combat-deepfake-election-content/ and an EU article on high-risk AI https://artificialintelligenceact.eu/article/6/ systems are not equivalent instruments. A single EU article can carry more legal force than a dozen state statutes combined, because U.S. state laws might be narrowly focused, and later repealed by a federal law. EU law, on the other hand, has supremacy over national laws of the member states. A single article might have consequences for entire industries. Comparing what each system prohibits outright is also instructive. Article 5 https://www.modulos.ai/blog/eu-ai-act-prohibited-practices/ of the EU AI Act bans eight categories of AI practices, including social scoring, real-time facial recognition in public spaces for law enforcement, emotion recognition in workplaces and schools, and pure profiling-based criminal prediction. These technology uses are clearly unacceptable, so category-wide bans make sense. However, four of these eight categories have use exceptions. For example, in emergency cases, law enforcement agencies are allowed to use facial recognition to find kidnapping victims. In comparison, the U.S. federal government has not banned any AI technology categories. Instead, during the previous presidency, federal agencies and then courts focused on a specific AI https://archives.hud.gov/news/2024/pr24-098.cfm use https://www.navigatehousing.com/fair-housing-guidance-withdrawal-what-huds-2026-notice-means-for-housing-providers/ , such as tenant screening and advertising housing opportunities, rather than banning the underlying tools. There is no AI without content access. So, in a broad sense, copyright laws are also part of AI regulation. The U.S. relies on the four-factor Fair Use doctrine https://www.copyright.gov/fair-use/ , which allows limited use of copyrighted material without permission. The EU, on the other hand, relies on a closed list of exceptions under the 2019 DSM Directive https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32019L0790 . The EU AI Act’s Article 4 https://artificialintelligenceact.eu/article/4/ permits commercial AI training on data only where rights holders haven’t actively blocked it. In practice, that means developers must audit their training data against a moving list of opt-outs. This problem does not exist on the other side of the Atlantic. Economists increasingly measure relevant regulation by what it adds to the price of building a product. For high-risk AI models, like the systems used in hiring https://www.fastcompany.com/section/hiring , credit scoring, or healthcare, EU AI Act compliance is estimated to increase in sectors with stringent validation requirements. In the U.S., the annual compliance burden is estimated to be between $50,000 and $150,000 https://www.kiteworks.com/cmmc-compliance/level-1-documentation-requirements/ . These costs are mostly related to data architecture for AI training and deployment and driven by state privacy laws like the California Consumer Privacy Act https://oag.ca.gov/privacy/ccpa . In other states without such comprehensive laws, the burden will still likely be driven by data handling regulation rather than AI laws directly. By that measure, in the EU, AI is roughly two to three times more expensive to launch in the high-risk tier, as defined by the EU AI Act https://ai-act-law.eu/ —AI systems that pose a significant risk of harm to health, safety, or fundamental individual rights. Small to medium-size enterprises face compliance costs ranging from €50,000 to €500,000 https://sqmagazine.co.uk/eu-ai-act-compliance-cost-statistics/ depending on use-case complexity, and that’s before the product has even shipped. This partially explains why between 2008 and 2021, close to 30% of the unicorns—privately held companies valued at $1 billion or more—that were founded in Europe relocated their headquarters https://www.insme.org/relocation-trends-of-eu-startups-and-scaleups-a-european-investment-bank-study/ outside the European Union, and why today U.S.-based AI companies attract roughly 75% of global AI venture capital https://renaissanceeurope.org/insights/europes-startup-problem-is-capital-scale-not-just-regulation , compared with just 6% for EU companies. Recently, Oxylabs became a unicorn https://oxylabs.io/blog/oxylabs-receives-investment , receiving a $130 million investment. We have no plans to relocate from Europe. But I understand why some firms do. The change in how the EU regulates AI is already underway. While no model is perfect, the EU is feeling greater pressure to reconsider its approach. The EU produces 22% of global https://knowledge.insead.edu/economics-finance/europes-historic-second-chance-leading-ais-next-wave AI research journal articles https://knowledge.insead.edu/economics-finance/europes-historic-second-chance-leading-ais-next-wave , compared to 17% by U.S.-based researchers. The problem is that research doesn’t translate into commercial leadership. EU policymakers are now saying so explicitly. In a May 2026 address https://ec.europa.eu/commission/presscorner/detail/en/speech 26 1159 to the Lisbon Council’s Scaling Europe Summit, Commissioner Dombrovskis acknowledged that excessive and overlapping regulatory obligations raise costs, deter investment, and divert resources from innovation. The Commission has since tabled https://ec.europa.eu/commission/presscorner/detail/en/statement 26 410 10 simplification https://ec.europa.eu/commission/presscorner/detail/en/statement 26 410 proposals targeting around €15 billion in annual savings for EU companies. It aims to keep the basic framework as is but make operating within the current framework less of a drag on commercial growth. It’s a start. Denas Grybauskas is chief governance and strategy officer at Oxylabs.