Article 42: Presumption of Conformity with Certain Requirements
Article 42 extends the presumption of conformity concept to two specific situations. First, high-risk AI systems that have been trained and tested on data reflecting the specific geographical, contextual, behavioural, or functional setting in which they are intended to be used are presumed to comply with the data governance requirements of Article 10(4). Second, high-risk AI systems that comply with the accessibility requirements under Directive (EU) 2019/882 (the European Accessibility Act) are presumed to be in conformity with the accessibility requirements of Article 13(2). This creates targeted presumptions linking AI Act compliance to existing Union law and good data practices.
Who does this apply to?
- -Providers using geographically and contextually appropriate training and testing data, who can leverage the Article 10(4) presumption of conformity
- -Providers leveraging European Accessibility Act (Directive 2019/882) compliance to satisfy the accessibility requirements of Article 13(2)
- -Conformity assessment bodies evaluating data governance and accessibility compliance of high-risk AI systems
Scenarios
A provider develops a high-risk AI system for road-traffic management intended for deployment in Nordic countries. The system is trained and tested on data that reflects Nordic traffic patterns, weather conditions, road infrastructure, and driving behaviours.
A provider of a high-risk AI-powered customer service kiosk demonstrates that its system meets all accessibility requirements under Directive (EU) 2019/882 (European Accessibility Act), including provisions for users with visual and hearing impairments.
Presumption 1: Data reflecting the intended setting (Article 10(4))
Article 10(4) requires that training, validation, and testing data sets take into account the specific geographical, contextual, behavioural, or functional setting within which the high-risk AI system is intended to be used.
Article 42 rewards providers who take this seriously: if your AI system has been trained and tested on data that genuinely reflects these characteristics, you benefit from a presumption of conformity with Article 10(4).
To rely on this presumption, providers should: - Document the data selection process, showing how datasets were curated to reflect the intended deployment context - Record the geographical, demographic, and contextual characteristics of the training and testing data - Demonstrate the match between data characteristics and the intended use environment - Retain versioned datasets or dataset cards for audit purposes
Presumption 2: European Accessibility Act compliance (Article 13(2))
Article 13(2) of the AI Act requires that high-risk AI systems meet accessibility requirements where they fall within the scope of accessibility obligations.
Article 42 creates a bridge to existing Union accessibility law: compliance with Directive (EU) 2019/882 — the European Accessibility Act (EAA) — is presumed to satisfy the AI Act's accessibility requirements under Article 13(2).
This means providers who already comply with the EAA do not need to undertake a separate accessibility assessment for AI Act purposes. The EAA covers products and services including: - Self-service terminals and kiosks - Consumer banking and payment services - E-commerce services - Electronic communications
Providers should document their EAA compliance and reference it in their AI Act conformity documentation.
Practical implications for conformity assessment
Both presumptions under Article 42 are rebuttable — market surveillance authorities can challenge them if evidence suggests the system does not actually meet the requirements despite the provider's claim.
For conformity assessment bodies, Article 42 simplifies assessment in two ways: 1. Where the provider demonstrates context-appropriate data, the body can accept the data governance presumption without a full independent data audit (though it may still verify the documentation) 2. Where EAA compliance is demonstrated, the body does not need to conduct a separate AI Act accessibility assessment
Providers should prepare documentation that clearly maps their claim to the Article 42 presumption and makes it easy for assessors to verify.
How Article 42 fits the conformity framework
- Article 10 — Data governance requirements, particularly Article 10(4) on context-appropriate data.
- Article 13 — Transparency and information requirements, including Article 13(2) on accessibility.
- Article 40 — Harmonised standards providing the general presumption of conformity.
- Article 43 — Conformity-assessment procedures where Article 42 presumptions may apply.
- Article 113 — Application dates.
Compliance checklist
- Document how your training and testing datasets reflect the geographical, contextual, behavioural, and functional setting of intended use.
- Maintain dataset cards or data documentation frameworks that record the provenance and characteristics of training data.
- If claiming the Article 10(4) presumption, ensure the match between data and deployment context is explicitly documented and auditable.
- If your system falls under the European Accessibility Act (Directive 2019/882), document your EAA compliance and reference it in your AI Act conformity file.
- Prepare clear mapping documents showing which Article 42 presumption you rely on and the supporting evidence.
- Retain all documentation for the lifetime of the system plus any required retention period for market surveillance purposes.
- Brief your conformity assessment body on your Article 42 claims before the formal assessment begins.
Check your data governance and accessibility compliance — start the free assessment.
Start Free AssessmentRelated Articles
Frequently asked questions
Does using geographically representative data automatically satisfy all of Article 10?
No. The presumption under Article 42 applies specifically to Article 10(4) — the requirement that data reflect the intended setting. Other Article 10 requirements (e.g., data quality, bias examination, privacy) must be demonstrated separately.
What if my system is deployed in a different context than it was trained for?
The presumption requires that training and testing data reflect the specific setting of intended use. If the deployment context changes significantly (e.g., from Nordic to Mediterranean traffic), the presumption no longer applies and the provider must reassess data governance compliance.
Is Directive 2019/882 compliance sufficient for all AI Act accessibility requirements?
Article 42 creates a presumption of conformity with Article 13(2) specifically. If the AI Act imposes accessibility-related requirements beyond Article 13(2) in other provisions, those would need to be addressed separately.