EU AI Act compliance guide: updates, scope, and key requirements

08/06/2026

    A practical guide to scope, timelines, standards, and sector impact.

    What the EU AI Act is and why it matters 

    The EU AI Act is the European Union’s legal framework for regulating certain systems based on risk. It sets out harmonized rules for the development, placing on the market, and use of those systems in the EU. 

    Its purpose is to support safe deployment, transparency, traceability, and respect for fundamental rights while allowing innovation to continue. 

    Unlike voluntary guidance, the Act: 

    • Applies across sectors. 
    • Covers both providers and deployers. 
    • Can apply to organizations outside the EU when systems are placed on the EU market or used in the EU. 
    • Includes conformity assessment, market surveillance, and penalties. 

    In practice, any organization developing, embedding, or using systems that fall within the Act’s scope may need to assess its obligations, regardless of where it is established. 

    EU AI Act risk categories 

    Risk classification is central to the regulation. The Act groups systems into categories with obligations that increase as potential impact increases. 

    Unacceptable risk

    Certain practices are prohibited because they are considered incompatible with fundamental rights or safety. Prohibited practices now include AI systems capable of generating sexualized deepfakes. Other examples include: 

    • Social scoring by public authorities. 
    • Certain manipulative or exploitative practices. 
    • Some biometric identification or categorization uses that are specifically restricted by the Act. 

    High-risk systems

    High‑risk classification under the Act depends on the role and function of the AI system within a given product or use case, rather than on the sector alone. 

    In practice, many high‑risk systems are found in regulated products or sensitive domains, including: 

    • Medical devices and in vitro diagnostics. 
    • Machinery and safety-related industrial systems. 
    • Automotive and transport systems. 
    • Employment, recruitment, and performance evaluation. 
    • Critical infrastructure. 

    These systems must meet requirements relating to: 

    • Risk management and data governance. 
    • Technical documentation. 
    • Human oversight. 
    • Accuracy, robustness, and cybersecurity. 
    • Post-market monitoring. 

    Scope of safety components 

    Under the AI Omnibus agreement, the definition of what qualifies as a safety component has been narrowed. 

    AI functions that solely assist users or optimise performance are not automatically classified as high‑risk, provided that their failure or malfunction does not create risks to health or safety. 

    As a result, high‑risk classification for AI systems embedded in regulated products depends on the system’s actual role and safety impact, rather than on the presence of AI alone. 

    Limited-risk systems 

    Some systems are mainly subject to transparency obligations. Common examples include: 

    • Chatbots. 
    • Generated or manipulated content, such as deepfakes, where such content does not fall under prohibited practices (sexualized) and is subject to transparency or watermarking obligations. 

    Minimal-risk systems 

    Most systems fall into this category and are not subject to additional obligations under the Act, although voluntary codes and internal controls may still be useful. 

    EU AI Act timeline

    The Act entered into force on 1 August 2024, and after the AI omnibus provisional agreement some original dates have been postponed, and obligations will apply progressively. 

    • 2 December 2026 - Watermarking / transparency obligations for AI‑generated content 
    • 2 December 2027 - Main requirements for high‑risk AI systems 
    • 2 August 2028 -  AI systems used as safety components and covered by EU sectoral safety & market surveillance legislation 

    For many organizations, December 2027 is now the key milestone for having governance, controls, and evidence ready. 

    Where ISO/IEC 42001 fits

    ISO/IEC 42001:2023 is the first international AI management system standard. It is voluntary, but it can support internal governance, documentation, and accountability.  

    ISO/IEC 42001 remains particularly relevant during the extended transition period under the AI Act, supporting governance, documentation, and accountability as organizations prepare for high‑risk obligations applying from December 2027. 

    The standard aligns well with themes that also appear in the Act, including: 

    • Risk management. 
    • Defined roles and responsibilities. 
    • Documentation. 
    • Oversight. 
    • Continuous improvement. 

    It is also compatible with other management systems, including ISO/IEC 27001. 

    In practice: 

    • The EU AI Act defines what must be achieved. 
    • ISO/IEC 42001 helps organizations structure how they organize and evidence it. 

    For high-risk systems, ISO/IEC 42001 can support many of the organizational controls needed for readiness, while EU-specific technical and conformity obligations still need to be addressed separately. 

    Related EU frameworks

    The EU AI Act does not stand alone. It sits within a broader regulatory environment. 

    Cyber Resilience Act 

    The Cyber Resilience Act focuses on the cybersecurity of products with digital elements. It overlaps with the AI Act in areas such as: 

    • Secure development. 
    • Vulnerability handling. 
    • Post-market surveillance. 

    This is especially relevant for AI-enabled products.

    GDPR and data governance 

    The Act complements GDPR, but it goes beyond personal data. High-risk systems introduce additional requirements around: 

    • Dataset quality. 
    • Bias mitigation. 
    • Traceability and record-keeping. 

    NIST AI RMF and international frameworks

    The NIST AI Risk Management Framework is not legally binding in the EU, but it can be useful for: 

    • Risk identification. 
    • Lifecycle-based assessments. 
    • Internal governance practices. 

    It is often used alongside ISO/IEC 42001 as a practical reference framework.

    ISO/IEC 27001

    ISO/IEC 27001 remains a strong foundation for information security. It supports readiness in areas such as: 

    • Security controls. 
    • Access control. 
    • Incident handling. 

    It is increasingly treated as a baseline for governance in organizations with mature compliance programs. 

    EU AI Act sector-specific impact

    Medical and healthcare

    Systems used in medical and healthcare settings are often high-risk by design. They also interact closely with MDR and IVDR requirements. 

    Key focus areas include: 

    • Clinical performance. 
    • Transparency and explainability. 
    • Post-market monitoring. 

    Automotive and transport 

    Systems in advanced driver assistance, autonomous functions, or safety components may fall within the high-risk framework and interact with existing vehicle rules. 

    Main priorities include: 

    • Safety validation. 
    • Human oversight. 
    • Cybersecurity and robustness.

    Industrial and machinery

    Systems used in industrial and machinery contexts may fall under both the EU AI Act and product‑specific legislation. However, not all AI systems in machinery are automatically classified as high‑risk. 

    Under the AI Omnibus agreement, the Machinery Regulation has been transferred to Annex I, Section B of the AI Act. The European Commission is expected to adopt delegated acts under the Machinery Regulation to specify AI‑related health and safety requirements, with application foreseen by August 2028. 

    In practice, only AI systems that qualify as safety components — meaning that their failure or malfunction could create risks to health or safety — are likely to trigger high‑risk obligations. 

    Employment, HR, and business processes 

    Systems used for recruitment, evaluation, or profiling are typically high-risk. 

    The main issues are: 

    • Bias and discrimination. 
    • Human review. 
    • Transparency toward individuals.

    From regulation to action

    Even before all harmonized standards are available, organizations can already take practical steps toward AI Act readiness. 

    The extended implementation timelines provide an opportunity for gradual preparation, including governance design, system inventory and classification, risk management procedures, and evidence collection to support future conformity assessments. 

    FAQs - Frequent asked questions 

    Is ISO/IEC 42001 mandatory under the Act? 

    No. ISO/IEC 42001 is voluntary, but it is widely recognized as a strong foundation for governance. 

    Does the Act apply to non-EU companies? 

    Yes, if their systems are placed on the EU market or used in the EU. 

    Are all systems regulated? 

    No. Some AI systems, including those embedded in regulated products, may fall outside high‑risk obligations where they do not qualify as safety components and do not create risks to health or safety. 

    Is this just another GDPR? 

    No. The Act focuses on system behavior, safety, and fundamental rights, not only on personal data. 

    Conclusion 

    Preparing for the EU AI Act is not only about meeting legal requirements. It is also about establishing clear governance, documentation, and control processes that support safe deployment across markets. 

    Disclaimer: This overview is based on the provisional AI Omnibus agreement. Final obligations and timelines may be refined following formal adoption and publication in the Official Journal of the European Union. 

     

     

    Applus+ uses first-party and third-party cookies for analytical purposes and to show you personalized advertising based on a profile drawn up based on your browsing habits (eg. visited websites). You can accept all cookies by pressing the "Accept" button or configure or reject their use. Consult our Cookies Policy for more information.

    Cookie settings panel