EU AI Act Article 6 high-risk obligations and Article 50 transparency rules are enforceable August 2, 2026, with penalties up to 35 million euros.
Regulation (EU) 2024/1689, the EU AI Act, became the first comprehensive binding legal framework for artificial intelligence in the world, and August 2, 2026 is the date on which its most commercially significant obligations become enforceable against every organization whose AI systems touch the European market, regardless of where that organization is incorporated or headquartered.
The Act’s enforcement structure is phased, and understanding which obligations are already in force is as important as preparing for those that begin in August 2026. Article 5’s prohibitions on unacceptable AI practices have been fully enforceable since February 2, 2025. Any organization that deployed a prohibited AI system after that date has already been in violation of binding EU law. Penalties for prohibited practice violations reach 35 million euros or seven percent of global annual turnover, whichever is the higher figure. The penalty structure is designed to be punishing even for large global enterprises: seven percent of global turnover for a company with $10 billion in annual revenue is $700 million. The scale of the penalty is calibrated to ensure that non-compliance is not an economically rational decision.
Article 5’s prohibited practices cover four categories. The first is AI systems that deploy subliminal or manipulative techniques to distort a person’s behavior causing significant harm, where the distortion occurs without the person’s awareness or consent. The second is AI systems that exploit vulnerabilities related to a person’s age, disability, or socio-economic circumstances to distort their behavior causing significant harm. The third is biometric categorization systems that infer or deduce sensitive attributes including race, political opinions, trade union membership, religious or philosophical beliefs, sex life, or sexual orientation. The fourth is social scoring systems used by public or private entities to evaluate or classify individuals based on social behavior or personal characteristics in ways that cause unjustified or disproportionate detriment. Any AI system that falls within these descriptions has been prohibited EU-market conduct since February 2, 2025.
August 2, 2026 triggers the high-risk AI system compliance obligations under Chapter III of the Act, incorporating Articles 9 through 15. Article 6 defines high-risk systems by reference to Annex I and Annex III. Annex III lists eight categories: biometrics, critical infrastructure, education and vocational training, employment and worker management, essential private and public services including credit scoring and health and life insurance pricing, law enforcement, migration and border control, and administration of justice and democratic processes. A company operating an AI-powered hiring tool, a credit scoring model, or a clinical decision support system is likely operating a high-risk AI system subject to the full Chapter III compliance architecture.
The Chapter III obligations for high-risk AI system providers are substantial and operationally demanding. Article 9 requires a risk management system: a documented, iterative process of identifying and analyzing known and reasonably foreseeable risks, adopting risk mitigation measures, and testing the residual risk level after mitigation. Article 10 imposes data and data administration requirements, including that training, validation, and testing datasets be subject to appropriate practices for data quality, relevance, representativeness, and error correction. Article 11 requires technical documentation prepared before market placement and maintained throughout the system’s life. Article 12 requires automatic logging of system events to enable post-hoc traceability of system behavior. Article 13 requires that deployers receive transparency documentation sufficient to allow informed human oversight. Article 14 requires that the system’s design enable effective human oversight during operation. Article 15 requires that the system achieve appropriate levels of accuracy, rigorousness, and cybersecurity.
Article 50’s transparency obligations also begin August 2, 2026 and apply more broadly than the high-risk category. Any AI system intended to interact directly with natural persons, including chatbots and voice assistants, must disclose to the person that they are interacting with an AI. Any organization that disseminates AI-generated content addressing matters of public interest must label that content as AI-generated. Violation of Article 50 carries penalties of up to 7.5 million euros or 1.5 percent of global annual turnover. For a US-based company using AI-powered customer service chatbots without disclosure to EU customers, Article 50 is the most immediately applicable August 2026 obligation.
The Act’s extra-territorial reach is explicit in Article 2. The Act applies to providers who place AI systems on the market or put them into service in the EU, regardless of whether the provider is established in the EU. It applies to deployers of AI systems located in the EU. It applies to providers and deployers located outside the EU when the output of the AI system is used in the EU. A US company that provides an AI-powered legal research tool to EU law firms, operates an AI-powered financial advisory service used by EU investors, or runs an AI-driven customer service system for EU customers is within the Act’s scope. The Act cannot be avoided by incorporating the AI provider entity outside the EU if the outputs of its systems are used within EU member states.
The conformity assessment process applies to high-risk AI systems before they are placed on the EU market. For most high-risk systems listed in Annex III, providers can conduct a self-assessment under Article 43. For certain high-risk systems involving biometrics and classified under Annex I, third-party conformity assessment by a notified body is required. Self-assessment requires preparation of the technical documentation under Article 11, verification that the system meets all Article 9 through 15 requirements, and registration of the system in the EU database for high-risk AI systems established under Article 71. Registration creates a public record of the system’s existence and intended purpose, enabling market surveillance authorities to identify and investigate non-compliant systems.
Providers must also establish a quality management system under Article 17 covering all aspects of the AI system’s lifecycle from design through post-market monitoring. Post-market monitoring is a continuing obligation under Article 72: providers must collect, document, and analyze data on their systems’ performance in real-world conditions, report serious incidents and malfunctions to national market surveillance authorities under Article 73, and implement corrective measures when the system fails to perform as documented. The post-market monitoring obligation transforms EU AI Act compliance from a pre-placement exercise into an ongoing operational commitment.
US companies assessing EU AI Act exposure should begin with a classification analysis. Not every AI system is high-risk, and not every AI application with a European nexus triggers the full Chapter III compliance architecture. The classification analysis asks: does the system fall within a category listed in Annex III, or does it serve as a safety component of a product covered by Annex I EU product safety legislation? If the answer to both questions is no, the system is likely a limited-risk or minimal-risk system subject only to Article 50 transparency obligations or no binding obligations at all. The classification analysis must be documented; regulators will ask for it in any market surveillance inquiry.
The unexpected implication for US companies is that EU AI Act compliance creates an organizational infrastructure that simultaneously satisfies several US regulatory expectations. The risk management documentation required under Article 9 maps directly onto the NIST AI RMF MANAGE function. The data regulatory management requirements of Article 10 align with data quality obligations referenced in SEC examination guidance for AI-using investment advisers. The human oversight requirements of Article 14 align with the human-in-the-loop validation standards that professional responsibility doctrine and enterprise risk management best practice both recommend. A company that builds its compliance program to meet EU AI Act requirements for its European operations will have a substantial portion of the documentation and operational infrastructure required for US regulatory purposes as well.
The EU AI Act’s general-purpose AI model provisions, which became applicable August 2, 2025, impose transparency and copyright compliance obligations on providers of large AI models made available to third parties. Providers of models trained using computational resources exceeding ten to the twenty-third power floating-point operations must comply with transparency requirements including technical documentation and compliance with the EU copyright framework for training data. Providers of models that present systemic risk, defined by reference to training compute and market reach thresholds, must conduct adversarial testing, report serious incidents to the AI Office, and maintain cybersecurity safeguards. US-based foundation model providers serving the EU market should have assessed these obligations when they became applicable in August 2025.
The practical compliance priority for an organization approaching August 2, 2026 is a three-step sequence: first, complete the classification analysis for every AI system with a European nexus; second, for every high-risk system identified, complete the Article 9 risk management documentation, Article 11 technical documentation, and Article 12 logging implementation before the enforcement date; and third, implement Article 50 transparency disclosures for every AI system that interacts directly with EU persons or generates public-interest content for EU distribution. Organizations that have not begun this sequence face a near-term compliance gap that will not be closed without a concentrated implementation effort in the weeks remaining before August 2, 2026.
What the EU AI Act ultimately represents for global enterprises is the emergence of a jurisdiction that has chosen to treat AI risk as a regulatory category equivalent to product safety and financial regulation, complete with mandatory pre-market assessment, ongoing monitoring obligations, and penalties calibrated to make non-compliance financially irrational. Whether the regulatory model achieves its stated safety objectives will be a matter for empirical evaluation over the coming years. That it creates enforceable legal obligations with significant financial consequences for organizations that ignore it is not a matter of prediction: it is the current state of EU law.
Interested in analysis about the intersection of corporate law, securities regulation, tech, artificial intelligence, public policy and the law? Check out my Substack channel. https://theinnovationattorney.substack.com/
The practical compliance priority for any organization approaching August 2, 2026 is a three-step sequence executed in the weeks remaining. First, complete the classification analysis for every AI system with a European nexus, documenting the analysis and the conclusion reached for each system. Second, for every high-risk system identified under Article 6, complete the Article 9 risk management documentation, implement the Article 12 logging architecture, and prepare the Article 11 technical documentation before the enforcement date. Third, implement Article 50 transparency disclosures for every AI system that interacts directly with EU persons, including chatbots, voice assistants, and automated content distribution systems. Organizations that have not begun this sequence face a compliance gap that requires concentrated implementation effort in the time available. The EU AI Act does not provide a grace period for organizations that were aware of the obligation and elected to defer preparation.