EU AI Act compliance: the complete guide

blockProhibited practices (Article 5)

These uses are deemed an unacceptable risk and are banned outright, with effect from 2 February 2025. Article 5

1. Subliminal, manipulative or deceptive techniques that materially distort behaviour and cause significant harm.
2. Exploiting vulnerabilities due to age, disability or socio-economic situation to distort behaviour and cause harm.
3. Social scoring that leads to detrimental or unjustified treatment.
4. Predictive policing based solely on profiling a person's risk of committing a crime.
5. Untargeted scraping of facial images from the internet or CCTV to build facial-recognition databases.
6. Emotion recognition in the workplace and in education (except medical or safety reasons).
7. Biometric categorisation inferring sensitive attributes such as race, political opinions or sexual orientation.
8. “Real-time” remote biometric identification in public spaces for law enforcement, with narrow, authorised exceptions.

warningHigh-risk AI: two routes in

An AI system is high-risk by one of two routes. Article 6

• Route 1, Annex I, safety components. AI that is a safety component of, or is itself, a product covered by existing EU product-safety law (machinery, medical devices, toys, lifts, radio equipment, motor vehicles, aviation) and required to undergo third-party conformity assessment.
• Route 2, Annex III, listed use cases. Stand-alone AI used in one of eight defined areas.

The eight Annex III areas are: Annex III

1. Biometrics (remote ID, categorisation, emotion recognition where not prohibited).
2. Critical infrastructure (safety components for traffic, water, gas, heating, electricity, digital infrastructure).
3. Education and vocational training (admissions, assessment, exam proctoring).
4. Employment and worker management (recruitment, screening, promotion, monitoring).
5. Access to essential services (credit scoring, public benefits, life and health insurance pricing, emergency triage).
6. Law enforcement (risk assessment, polygraphs, evidence evaluation, profiling).
7. Migration, asylum and border control (risk assessment, application examination, person detection).
8. Administration of justice and democratic processes (assisting judges, influencing elections).

visibilityLimited risk: the transparency duties (Article 50)

Some AI is not high-risk but still needs to be honest about what it is. Article 50

• Chatbots: people must be told they are interacting with an AI, unless it is obvious.
• AI-generated content: providers must mark synthetic audio, image, video or text in a machine-readable way.
• Deepfakes: deployers must disclose that content is artificially generated or manipulated.
• Emotion recognition and biometric categorisation: deployers must inform the people exposed to it.

groupsWho must comply, and the extraterritorial reach

The duties fall mainly on two roles, with importers, distributors and product manufacturers also caught. Articles 16 and 26

A deployer, importer or distributor can be reclassified as a provider (taking on full provider duties) if it puts its name on a high-risk system, substantially modifies it, or changes its intended purpose. Article 25 See the full breakdown in the provider vs deployer obligations guide.

hubGeneral-purpose AI (GPAI) models

Broadly capable models, such as large language models, are regulated on their own track under Chapter V, in force since 2 August 2025. The Commission's guidelines use an indicative marker of more than 10²³ FLOP of training compute to identify a GPAI model. GPAI guidelines

All GPAI model providers (Article 53) must keep technical documentation, give information to downstream providers who integrate the model, have a policy to comply with EU copyright law, and publish a summary of the content used for training. Free and open-source models are exempt from some documentation duties, unless they carry systemic risk. Article 53

GPAI models with systemic risk (Article 55), presumed when training compute exceeds 10²⁵ FLOP, face extra duties even when open-source: model evaluation including adversarial testing (red-teaming), assessing and mitigating systemic risks, reporting serious incidents to the AI Office, and adequate cybersecurity. Article 55

The voluntary GPAI Code of Practice, drawn up under the AI Office, helps providers demonstrate compliance with Articles 53 and 55 while harmonised standards are still being finalised. Code of Practice

scheduleThe timeline at a glance

• 1 Aug 2024: the Act enters into force.
• 2 Feb 2025 (in force): Article 5 prohibitions and Article 4 AI literacy duty apply.
• 2 Aug 2025 (in force): GPAI model rules, governance (the AI Office) and penalties apply.
• 2 Aug 2026: most high-risk (Annex III) and Article 50 transparency obligations apply.
• 2 Aug 2027: high-risk AI that is a safety component of regulated (Annex I) products, plus GPAI models placed on the market before 2 Aug 2025.

See the living timeline tracker for the proposed Digital Omnibus changes and the current status of each date. Implementation timeline

gavelPenalties

Fines are tiered, and the regime has been in force since 2 August 2025. Article 99

• Prohibited practices (Article 5): up to €35 million or 7% of total worldwide annual turnover, whichever is higher.
• Other obligations (including high-risk and transparency): up to €15 million or 3%.
• Misleading information to authorities: up to €7.5 million or 1%.

For SMEs and start-ups, each fine is capped at the lower of the percentage or the fixed amount, and authorities must weigh proportionality.

schoolAI literacy (Article 4)

In force since 2 February 2025, this is a duty most organisations can act on now. Providers and deployers must take measures to ensure, to their best extent, a sufficient level of AI literacy among staff and others operating AI on their behalf, tailored to their knowledge and the context of use. It applies to all AI systems, not just high-risk, and to non-EU entities in scope. Article 4 See the AI literacy guide.

verified_userISO 42001 and harmonised standards

ISO/IEC 42001 is the international standard for an AI management system, and a strong head-start on the governance controls the Act expects. But on its own it is an international standard, not a harmonised European one, and so confers no automatic presumption of conformity with the Act. ISMS.online

That presumption comes from following harmonised standards whose references are published in the Official Journal. CEN-CENELEC is still developing them. Until they land, ISO/IEC 42001 and the GPAI Code of Practice are practical bridges, not legal shields. Article 40 See our AI governance guide.