🇪🇺Artificial Intelligence Act

← Back to laws in European Union

The Artificial Intelligence Act regulates AI through a risk-based framework

The Artificial Intelligence Act is the European Union's horizontal legal framework for artificial intelligence. It does not treat every AI system in the same way. Instead, it uses a risk-based model: some AI practices are prohibited, some AI systems are classified as high-risk and subject to detailed obligations, some AI systems are subject mainly to transparency duties, and general-purpose AI models have their own compliance layer.

The Act is important for privacy and data governance because AI systems often rely on personal data, biometric data, profiling, automated decision-making, training datasets, monitoring tools and large-scale inference. The Act does not replace the GDPR. It sits beside data protection law and adds AI-specific duties around risk management, human oversight, transparency, documentation, accuracy, robustness, cybersecurity and governance.

What the Act is trying to achieve

Article 1 sets the purpose of the Act. It aims to improve the functioning of the internal market by creating a uniform legal framework for the development, placing on the market, putting into service and use of AI systems in the Union. The law promotes human-centric and trustworthy AI while protecting health, safety, fundamental rights, democracy, the rule of law and environmental protection.

The practical idea is simple: AI should be allowed to innovate, but not at the cost of opaque, unsafe or discriminatory systems being deployed in sensitive contexts. The Act therefore focuses heavily on the role of providers, deployers, importers, distributors and other operators in the AI value chain.

Who needs to pay attention

The Act applies to a wide range of actors. Providers that develop or place AI systems or general-purpose AI models on the EU market are central to the framework. Deployers that use AI systems under their authority also have important duties, especially where they use high-risk AI systems. Importers, distributors, product manufacturers and authorised representatives may also have obligations depending on their role.

The Act is not limited to companies physically established in the EU. It can also apply to providers and deployers established in third countries where the output produced by an AI system is intended to be used in the Union. This matters for non-EU businesses that sell AI products into the EU, provide AI-enabled services to EU customers, or supply AI outputs used in EU decision-making.

What counts as an AI system

Article 3 defines an AI system as a machine-based system designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment. For explicit or implicit objectives, it infers from the input it receives how to generate outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments.

This definition is broad enough to cover modern AI systems while excluding ordinary software that merely follows rules set by humans without the relevant capacity to infer. For compliance work, the first step is therefore not to ask whether a tool is branded as AI. The better question is whether the system uses machine-based inference to generate outputs that influence decisions, content, recommendations, predictions or environments.

Prohibited AI practices

Article 5 prohibits AI practices considered unacceptable. These include manipulative or deceptive AI techniques that materially distort behaviour and cause or are likely to cause significant harm, exploitative systems targeting vulnerabilities related to age, disability or social or economic situation, certain social scoring systems, and certain predictive policing practices based solely on profiling or personality traits.

The Act also prohibits specific biometric practices, including certain biometric categorisation systems that infer sensitive characteristics, untargeted scraping of facial images from the internet or CCTV footage to create or expand facial recognition databases, and emotion recognition systems in workplace and educational contexts except where used for medical or safety reasons. Real-time remote biometric identification in publicly accessible spaces for law enforcement is generally prohibited, subject only to narrow exceptions and strict safeguards.

High-risk AI systems

The most detailed part of the Act applies to high-risk AI systems. Article 6 identifies two routes into the high-risk category. First, an AI system is high-risk where it is a safety component of a product, or is itself a product, covered by specified EU harmonisation legislation and the product requires third-party conformity assessment. Second, AI systems listed in Annex III are treated as high-risk unless a limited exception applies because the system does not pose a significant risk of harm to health, safety or fundamental rights.

Annex III covers sensitive use areas such as biometrics, critical infrastructure, education and vocational training, employment and worker management, access to essential private and public services, law enforcement, migration, asylum and border control management, and administration of justice and democratic processes. These are areas where AI outputs can shape life chances, legal position, safety, liberty, access to services or fundamental rights.

Core requirements for high-risk AI systems

High-risk AI systems must meet a set of technical and organisational requirements before they are placed on the market or put into service. These requirements are not cosmetic. They are intended to make the system auditable, explainable enough for its purpose, technically controlled, and safer for affected persons.

Articles 8 to 15 cover risk management, data and data governance, technical documentation, record-keeping, transparency and provision of information to deployers, human oversight, accuracy, robustness and cybersecurity. In practical terms, providers of high-risk AI systems need to build a compliance file around the system: intended purpose, foreseeable misuse, risk controls, training and testing data governance, instructions for use, logging, performance metrics, oversight measures and post-market monitoring.

Risk management and data governance

Article 9 requires a risk management system for high-risk AI systems. This means identifying, estimating and evaluating risks, adopting risk-control measures, and testing whether those measures work. The risk management process must consider the intended purpose of the system and reasonably foreseeable misuse.

Article 10 deals with data and data governance. Where high-risk AI systems use data for training, validation or testing, the datasets must be subject to appropriate data governance and management practices. This is particularly important where poor-quality, biased, incomplete or unrepresentative data can produce discriminatory or unsafe outcomes.

Documentation, logging and transparency

Article 11 requires technical documentation for high-risk AI systems. Article 12 requires record-keeping capabilities through logs where appropriate. Article 13 requires high-risk AI systems to be designed and developed in a sufficiently transparent way so that deployers can interpret the system's output and use it appropriately.

This does not mean that every model must reveal all trade secrets or source code to every user. It does mean that the system must be accompanied by enough information for compliant deployment, meaningful human oversight, risk management and accountability. For high-risk systems, vague vendor claims such as “AI-powered” or “fully automated intelligence” are not enough.

Human oversight, accuracy and cybersecurity

Article 14 requires high-risk AI systems to be designed and developed so that they can be effectively overseen by natural persons during the period in which the system is in use. Human oversight is intended to prevent or minimise risks to health, safety or fundamental rights.

Article 15 requires high-risk AI systems to achieve an appropriate level of accuracy, robustness and cybersecurity throughout their lifecycle. The expected level of performance must be declared in the accompanying instructions. This is important because AI compliance cannot stop at launch. Systems may drift, degrade, face adversarial attacks, or behave unexpectedly in real-world environments.

Obligations of providers

Providers of high-risk AI systems carry the heaviest compliance burden. They must ensure that the system complies with the Act's requirements, implement a quality management system, prepare technical documentation, keep logs where under their control, undergo conformity assessment where required, draw up an EU declaration of conformity, affix CE marking where applicable, and take corrective action where the system is not compliant.

Providers must also maintain post-market monitoring and report serious incidents in the manner required by the Act. For AI governance teams, this means provider compliance should be managed across the product lifecycle: design, training, testing, deployment, monitoring, incident response, updates and withdrawal where necessary.

Obligations of deployers

Deployers are not passive users. Where they use high-risk AI systems, they must take appropriate technical and organisational measures to use the system in accordance with the instructions for use. They must assign human oversight to persons with appropriate competence, training, authority and AI literacy, monitor the system's operation, and keep logs where those logs are under their control.

Deployers using high-risk AI systems in workplace contexts must also inform workers' representatives and affected workers before the system is put into service or used, where relevant under the Act. Deployers using certain high-risk AI systems to make or assist decisions about natural persons may also need to inform those persons that they are subject to the use of the system.

Transparency obligations for certain AI systems

Article 50 creates transparency obligations for specific AI systems, even where they are not necessarily high-risk. People should be informed when they are interacting with an AI system unless this is obvious from the circumstances and context of use. Deployers of emotion recognition systems or biometric categorisation systems must inform exposed natural persons of the operation of the system, subject to the Act's conditions and exceptions.

The Act also addresses synthetic or manipulated content. AI-generated or manipulated image, audio or video content that constitutes a deep fake must generally be disclosed as artificially generated or manipulated. This is important for media, political communication, advertising, education, entertainment and fraud prevention.

General-purpose AI models

The Act creates a separate layer for general-purpose AI models. These are models capable of performing a wide range of distinct tasks and capable of being integrated into a variety of downstream systems or applications. Providers of general-purpose AI models must prepare and keep technical documentation, provide information to downstream providers that integrate the model, put in place a policy to comply with EU copyright law, and publish a sufficiently detailed summary of the content used for training according to the template provided by the AI Office.

Open-source general-purpose AI models receive some limited relief from certain obligations where the required conditions are met. However, that relief does not apply to general-purpose AI models with systemic risk. Providers of models with systemic risk must meet additional obligations, including model evaluation, adversarial testing, systemic risk assessment and mitigation, serious incident reporting and adequate cybersecurity protection.

AI literacy and governance culture

The Act also recognises AI literacy. Organisations involved in the AI value chain should ensure that people dealing with AI systems have the necessary understanding to use, oversee and make decisions about those systems responsibly. This is not just a training issue. It affects procurement, product design, HR, legal review, compliance, cybersecurity, data governance and senior management oversight.

For an organisation, AI literacy should mean that relevant teams understand what the system does, what it does not do, what data it uses, where it may fail, what human checks are required, how outputs should be interpreted, and what to do when the system behaves unexpectedly.

Relationship with GDPR and data protection law

The Artificial Intelligence Act does not replace the GDPR, the Law Enforcement Directive, the ePrivacy framework or other EU data protection rules. Where an AI system processes personal data, data protection law continues to apply. The Act expressly sits beside those laws and does not remove obligations relating to lawful basis, transparency, data subject rights, automated decision-making, DPIAs, security, processor contracts or international transfers.

This relationship is especially important for biometric identification, biometric categorisation, emotion recognition, profiling, recruitment tools, credit scoring, insurance pricing, health AI, education technology and worker monitoring. In these areas, an organisation may need both an AI Act analysis and a GDPR analysis. Passing one does not automatically satisfy the other.

Innovation measures and regulatory sandboxes

The Act also contains measures to support innovation. Member States must ensure that at least one AI regulatory sandbox is established at national level. These sandboxes are intended to provide a controlled environment for developing, training, testing and validating innovative AI systems before they are placed on the market or put into service.

The innovation chapter is important because the Act is not only a restriction framework. It is also designed to support safe innovation, especially for SMEs and start-ups, by creating supervised pathways for experimentation and compliance learning.

Supervision and enforcement

The Act creates a governance structure involving national competent authorities, market surveillance authorities, the European Artificial Intelligence Board and the AI Office. The AI Office has a particularly important role in relation to general-purpose AI models.

Article 99 sets the framework for penalties. Member States must lay down penalties that are effective, proportionate and dissuasive. The Act also sets maximum administrative fines for certain categories of infringement, including higher maximums for prohibited AI practices and lower tiers for other infringements or supplying incorrect, incomplete or misleading information to authorities.

Key application dates

The Act entered into force after publication in the Official Journal. Article 113 states that it applies from 2 August 2026, but some parts apply earlier. Chapters I and II apply from 2 February 2025. Several provisions, including the general-purpose AI chapter, governance provisions and penalties, apply from 2 August 2025, with specific exceptions. Article 6(1) and corresponding obligations apply from 2 August 2027.

These staggered dates matter. Organisations should not wait for the general application date to begin AI governance. Prohibited practices, AI literacy, general-purpose AI obligations, governance structures and high-risk system preparation have different timelines and should be mapped separately.