Article 84: Union AI Testing Support Structures
Article 84 empowers the Commission to designate one or more Union AI testing support structures to assist with the implementation and enforcement of the AI Act. These structures provide independent technical expertise, testing capabilities, and benchmarks for evaluating AI systems. They support national market surveillance authorities in their enforcement tasks — particularly when evaluating complex AI systems under Articles 79–82. The Commission may designate existing Union bodies (such as the Joint Research Centre — JRC) or establish new structures. Testing support structures also contribute to standardisation efforts, develop testing methodologies, and publish guidance. This ensures that enforcement authorities across the Union have access to state-of-the-art technical capacity rather than relying solely on national resources. Always verify on EUR-Lex.
Who does this apply to?
- -The European Commission, which designates and oversees Union AI testing support structures
- -Existing Union bodies such as the Joint Research Centre (JRC) that may be designated as testing support structures
- -National market surveillance authorities that rely on these structures for independent technical expertise and testing support
Scenarios
A national market surveillance authority in Spain needs to evaluate whether a complex generative AI system integrated into a high-risk medical device complies with Article 15 accuracy and robustness requirements. The authority lacks in-house expertise on large language model evaluation and adversarial testing.
The Commission designates the JRC as a Union AI testing support structure. Multiple Member State authorities have inconsistent approaches to evaluating bias in Annex III employment AI systems. The testing structure develops a standardised bias testing methodology and publishes it as guidance.
What Article 84 does (plain language)
Enforcing AI regulation requires deep technical expertise — understanding model architectures, testing for bias, evaluating robustness against adversarial inputs, assessing data quality, and benchmarking accuracy. Most national market surveillance authorities were not built for this. Article 84 solves this by creating centralised technical support:
- The Commission may designate one or more Union AI testing support structures — dedicated bodies that provide technical services to enforcement authorities
- These structures offer independent testing of AI systems, develop benchmarks and testing methodologies, and provide technical advice to national authorities
- They may also contribute to standardisation efforts, helping to develop the harmonised standards that define compliance
- The Commission may designate existing bodies (the JRC is the most likely candidate) or create new structures as needed
- The structures operate independently — their technical findings are not influenced by the providers whose systems they evaluate
Tasks of Union AI testing support structures
Article 84 testing support structures perform several key functions:
Technical testing and evaluation: Provide hands-on testing of AI systems at the request of market surveillance authorities — including accuracy benchmarking, bias auditing, robustness testing, and cybersecurity assessment.
Methodology development: Develop and publish standardised testing methodologies that national authorities can adopt, ensuring consistent enforcement across the Union.
Standardisation support: Contribute to the development of harmonised standards under Article 40, providing technical input to European Standardisation Organisations (CEN, CENELEC, ETSI).
Capacity building: Help national authorities build their in-house expertise through training, guidance, and technical documentation.
Benchmarking: Create and maintain benchmarks for evaluating AI system performance across different domains and risk categories.
Advisory role: Provide the Commission, the AI Office, and the AI Board with independent technical advice on enforcement matters and regulatory implementation.
How Article 84 connects to the rest of the Act
- Article 74 — Market surveillance and control framework that the testing structures support.
- Article 79 — National enforcement procedure where authorities may need testing support to evaluate risky systems.
- Article 80 — Classification challenges where technical evaluation of the system is required.
- Article 40 — Harmonised standards to which testing structures contribute.
- Article 64 — AI Office responsibilities that testing structures complement.
- Article 113 — Application dates; Article 84 applies from 2 August 2026.
Official wording (excerpt): Article 84
Editorial note: The full authentic text of Article 84 is published on EUR-Lex. The following is a faithful summary of its core operative provisions.
The Commission shall designate one or more Union AI testing support structures to perform tasks listed in this Article in the area of artificial intelligence. Those testing support structures shall advise and assist market surveillance authorities in carrying out their tasks under this Regulation, including by providing independent technical expertise and by performing testing of AI systems. The testing support structures shall also advise and assist the AI Office in the execution of its tasks. Union AI testing support structures shall carry out their activities with independence and impartiality. They shall contribute to the development of benchmarks and testing methodologies and make their findings publicly available to the extent possible.
Compliance checklist
- Understand that your AI system may be evaluated by a Union testing support structure — not just the national authority — especially for complex technical assessments.
- Prepare comprehensive technical documentation (Article 11) that enables independent third-party testing of your system's accuracy, robustness, and bias characteristics.
- Monitor published testing methodologies and benchmarks from designated structures — they signal the technical standards authorities will use in enforcement.
- If a testing structure identifies deficiencies in your system during an Article 79/80 evaluation, prepare to respond with technical evidence and corrective action plans.
- Track which Union bodies are designated as testing support structures — their published guidance and methodologies become de facto enforcement standards.
- Ensure your system's APIs or testing interfaces can support independent evaluation — authorities backed by testing structures may request structured access for benchmarking.
Understand enforcement testing—start the free assessment.
Start Free AssessmentRelated Articles
Article 74: Market Surveillance and Control of AI Systems in the Union Market
Article 79: Procedure at National Level for AI Systems Presenting a Risk
Article 80: Procedure for Dealing with AI Systems Classified by the Provider as Non-High-Risk in Application of Annex III
Article 113: Entry into Force and Application Dates
Frequently asked questions
Will the JRC be the main Union AI testing support structure?
The JRC (Joint Research Centre) is widely expected to play a central role, given its existing expertise in AI evaluation and its mandate as the Commission's in-house science and knowledge service. However, Article 84 allows the Commission to designate multiple structures, and other bodies or newly created entities may also be designated.
Can a testing support structure directly enforce the AI Act against providers?
No. Testing support structures provide technical support and independent evaluation — they advise and assist national market surveillance authorities and the AI Office. Enforcement decisions (restrictions, prohibitions, withdrawals, recalls) remain the competence of national authorities. The testing structure's findings inform enforcement but do not constitute enforcement actions themselves.
Do testing support structures publish their methodologies?
Yes. Article 84 requires these structures to make their findings publicly available to the extent possible. Published benchmarks and testing methodologies effectively become reference standards for the industry, giving providers advance insight into the technical criteria authorities will apply.