Recitals

On this page, you can find recitals provided under the EU AI Act’s official text.

The 180 recitals are the legislator’s stated reasoning. They do not impose obligations of their own, but courts and regulators use them to interpret the articles that do, which makes them the place to look when an article’s wording is capable of more than one reading. Each recital below links to its EUR-Lex source.

• Recital 1Purpose and the internal market
• Recital 2Union values and the Charter
• Recital 3Divergent national rules
• Recital 4The benefits of AI
• Recital 5AI risks and harms
• Recital 6Human-centric AI
• Recital 7Why common high-risk rules
• Recital 8Why a Union framework
• Recital 9The New Legislative Framework
• Recital 10Data protection law
• Recital 11The Digital Services Act
• Recital 12Defining an AI system
• Recital 13Defining a deployer
• Recital 14Defining biometric data
• Recital 15Biometric identification defined
• Recital 16Biometric categorisation defined
• Recital 17Remote biometric identification
• Recital 18Emotion recognition defined
• Recital 19Publicly accessible space
• Recital 20The AI literacy duty
• Recital 21Providers outside the Union
• Recital 22Extraterritorial effect
• Recital 23Union institutions and bodies
• Recital 24Military and national security
• Recital 25Research and development
• Recital 26The risk-based approach
• Recital 27Trustworthy AI ethics guidelines
• Recital 28Manipulation and social control
• Recital 29Deceptive techniques
• Recital 30Categorising sensitive traits
• Recital 31Social scoring
• Recital 32Real-time biometric identification
• Recital 33Narrow biometric exceptions
• Recital 34Safeguards on authorised use
• Recital 35Judicial authorisation
• Recital 36Notifying the authorities
• Recital 37National implementation
• Recital 38Biometric data in policing
• Recital 39Other biometric data processing
• Recital 40Ireland and Protocol No 21
• Recital 41Denmark and Protocol No 22
• Recital 42Presumption of innocence
• Recital 43Untargeted facial scraping
• Recital 44The science of emotion inference
• Recital 45Practices banned by other law
• Recital 46High-risk requirements
• Recital 47Impact on health and safety
• Recital 48Impact on fundamental rights
• Recital 49Transport and aviation law
• Recital 50Annex I products
• Recital 51Classification and product safety
• Recital 52Stand-alone high-risk systems
• Recital 53The no-significant-risk derogation
• Recital 54Biometric use cases
• Recital 55Critical infrastructure
• Recital 56Education and vocational training
• Recital 57Employment and worker management
• Recital 58Essential services and benefits
• Recital 59Law enforcement
• Recital 60Migration, asylum and borders
• Recital 61Justice and democratic processes
• Recital 62Elections and voting behaviour
• Recital 63Classification and other law
• Recital 64Requirements and the state of the art
• Recital 65The risk management system
• Recital 66Scope of the requirements
• Recital 67Data quality and governance
• Recital 68Testing and experimentation
• Recital 69Data protection by design
• Recital 70Special categories and bias
• Recital 71Traceability and documentation
• Recital 72Transparency for deployers
• Recital 73Human oversight
• Recital 74Accuracy and robustness
• Recital 75Technical robustness
• Recital 76Cybersecurity
• Recital 77The Cyber Resilience Act
• Recital 78Cybersecurity in assessment
• Recital 79The provider's responsibility
• Recital 80Accessibility and disability
• Recital 81Quality management
• Recital 82Authorised representatives
• Recital 83Certainty in the value chain
• Recital 84Becoming the provider
• Recital 85General-purpose as high-risk
• Recital 86Change of provider
• Recital 87Product manufacturers
• Recital 88Suppliers of tools and components
• Recital 89Third-party components
• Recital 90Model contractual terms
• Recital 91Importers and distributors
• Recital 92Worker information rights
• Recital 93The deployer's role
• Recital 94Biometric identification in policing
• Recital 95Post-remote identification
• Recital 96Fundamental rights impact assessment
• Recital 97Defining a general-purpose model
• Recital 98The generality threshold
• Recital 99Large generative models
• Recital 100Becoming general-purpose
• Recital 101Model providers in the chain
• Recital 102Free and open-source licences
• Recital 103Open-source AI components
• Recital 104Open-source model providers
• Recital 105Copyright and generative models
• Recital 106Model provider duties
• Recital 107Training-content summary
• Recital 108Copyright policy
• Recital 109Proportionate model obligations
• Recital 110What systemic risk means
• Recital 111Classifying systemic risk
• Recital 112The classification procedure
• Recital 113Reclassification
• Recital 114Systemic-risk model duties
• Recital 115Mitigating systemic risk
• Recital 116Drawing up codes of practice
• Recital 117Codes as the compliance route
• Recital 118Scope and other Union law
• Recital 119Digital services instruments
• Recital 120Disclosing generated content
• Recital 121Standardisation
• Recital 122Presumption of conformity
• Recital 123Assessment before market entry
• Recital 124Avoiding duplicate assessment
• Recital 125Notified body involvement
• Recital 126Notifying assessment bodies
• Recital 127Mutual recognition
• Recital 128Substantial modification
• Recital 129CE marking
• Recital 130Rapid access to innovation
• Recital 131The EU database
• Recital 132Impersonation and deception
• Recital 133Marking synthetic content
• Recital 134Deployer deep-fake disclosure
• Recital 135Transparency codes of practice
• Recital 136Disclosure and other Union law
• Recital 137Transparency and other law
• Recital 138Controlled experimentation
• Recital 139Sandbox objectives
• Recital 140Personal data in a sandbox
• Recital 141Real-world testing
• Recital 142Beneficial AI
• Recital 143SMEs and start-ups
• Recital 144The AI-on-demand platform
• Recital 145Knowledge and expertise
• Recital 146Microenterprises
• Recital 147Access to testing facilities
• Recital 148The governance framework
• Recital 149The European AI Board
• Recital 150The advisory forum
• Recital 151The scientific panel
• Recital 152AI testing support structures
• Recital 153National competent authorities
• Recital 154Independence of authorities
• Recital 155Learning from monitoring
• Recital 156The market surveillance system
• Recital 157Fundamental rights bodies
• Recital 158Financial services supervision
• Recital 159Surveillance of biometrics
• Recital 160Joint investigations
• Recital 161Competences for built-on systems
• Recital 162Centralised model supervision
• Recital 163The panel and model monitoring
• Recital 164AI Office enforcement
• Recital 165Voluntary application
• Recital 166Codes for non-high-risk products
• Recital 167Confidentiality
• Recital 168Penalties and enforcement
• Recital 169Fines on model providers
• Recital 170The right to complain
• Recital 171The right to an explanation
• Recital 172Whistleblower protection
• Recital 173Delegated acts
• Recital 174Evaluation and review
• Recital 175Implementing acts
• Recital 176Subsidiarity and proportionality
• Recital 177Transitional arrangements
• Recital 178Voluntary early compliance
• Recital 179Application dates
• Recital 180The EDPS and EDPB opinion