Article 60 allows providers or prospective providers of high-risk AI systems listed in Annex III to test those systems in real world conditions outside an AI regulatory sandbox, subject to a testing plan, safeguards and, where the Act requires it, approval by the market surveillance authority. The provider or prospective provider must register specified information in the EU database under Annex IX before testing, keep it up to date, and respect the conditions on test subjects, data protection and suspension or termination. Testing in real world conditions is a supervised route to evidence, not an unregulated deployment, and it must be distinguished from placing a system on the market.
What is testing in real world conditions?
Testing in real world conditions means testing a high-risk AI system listed in Annex III in a real environment, with real users or real data, before the system is placed on the market or put into service, in order to gather evidence about its performance, safety and compliance. It differs from laboratory testing because the environment, the users and the consequences are real, and it differs from deployment because the system is not yet being supplied as a finished product. The Act treats that difference as a reason for supervision rather than as a reason for freedom: the test must follow a plan, respect safeguards, and be registered and, where required, approved. A team that describes a live deployment as a test, or that tests without a plan in order to avoid conformity work, misunderstands the provision. The test is a defined, time-limited, documented activity with a named purpose, and its results feed into the conformity and documentation work that follows.
What must a testing plan contain?
A testing plan must set out the purpose of the test, the system and version to be tested, the test subjects and environment, the duration and scale, the safeguards, the data to be collected, the criteria for success and failure, and the arrangements for suspension or termination. The plan should also address data protection, including the legal basis for processing personal data of test subjects, and should provide for informed consent where the Act requires it, with clear information about the test and the right to withdraw. Test subjects must be informed, and vulnerable persons require particular protection. The plan is not a marketing document: it should be written so that a market surveillance authority can assess, from the plan alone, whether the test is proportionate, whether the safeguards are adequate and whether the evidence it will produce can support a later conformity decision. A plan that cannot answer those questions should not be submitted.
For the broader regulatory framework, see EU database, registration and post-market monitoring.
What approval and registration are required?
Before testing, the provider or prospective provider must submit the testing plan to the market surveillance authority, and the test may proceed only in accordance with the approval arrangements in Article 60, which include deemed approval after a specified period where the authority does not respond, subject to the conditions in the provision. The provider or prospective provider must also register the information listed in Annex IX in the EU database, including a Union-wide unique identification number for the test, the names and contact details of the provider or prospective provider and the deployers involved, a brief description of the system and its intended purpose, a summary of the main characteristics of the testing plan, and information on any suspension or termination. That registration must be kept up to date throughout the test, so every material change to the plan is logged as it happens. The registration information is restricted to market surveillance authorities and the Commission unless consent is given for public access.
What safeguards apply to test subjects?
Test subjects are people, not data points, and the Act sets safeguards that reflect that. Subjects must be informed about the test, its purpose and their participation, and must give informed consent where required, with the ability to withdraw. Particular protection is required for vulnerable persons, and the test must not expose subjects to risks beyond what the plan justifies and the authority has accepted. Personal data collected during testing must be processed lawfully, with appropriate safeguards, and must not be used for purposes beyond the test without a proper basis. The provider should also plan for what happens to a subject who is harmed or disadvantaged during the test, including how the test will be suspended, how the subject will be supported and how the incident will be recorded. Those arrangements should be in the plan before testing begins, because a safeguard improvised after harm has occurred is not a safeguard.
How does testing connect to sandboxes and conformity?
Testing in real world conditions under Article 60 sits alongside, not instead of, AI regulatory sandboxes under Articles 57 to 59. A sandbox provides a controlled environment supervised by a competent authority, with a sandbox plan and an exit report, while Article 60 testing can take place outside a sandbox subject to its own plan and approval. The two routes can complement each other: sandbox work can inform a real-world test plan, and real-world test evidence can inform conformity assessment. In either case, the evidence gathered must be fed into the technical documentation, risk management and conformity record, and a successful test does not by itself establish compliance. The provider should record which route was used, which plan governed it, what evidence was gathered and what conclusions were drawn, so a later reviewer can see how the test evidence supports, or fails to support, the conformity decision.
Continue with what is public and what is restricted in the EU database.
What does this guide not decide?
This guide does not decide whether a particular test is permitted, whether approval is required or deemed, or whether the safeguards for a given test are adequate. Those questions depend on the system, the test design, the subjects and the Member State concerned, and they require Article 60 and Annex IX, together with data protection law and any national implementation rules, to be checked for the specific case. The guide also does not cover testing of Annex I product systems, for which separate provision has been made, and that position should be checked in the consolidated text before any reliance. Where a real-world test is planned, confirm the approval, registration and safeguard position with the relevant market surveillance authority and legal adviser before any subject is involved.
Frequently asked questions
Is real-world testing the same as launching a product?
No. Testing in real world conditions is a defined, time-limited, supervised activity under a testing plan, with registration and approval arrangements. Placing a system on the market is a separate step that requires conformity work and registration under Article 49.
What must be registered before testing?
The Annex IX information, including the test identification number, the provider and deployers involved, a brief description of the system and its purpose, a summary of the testing plan, and information on suspension or termination, kept up to date throughout the test.
Must test subjects consent?
Subjects must be informed about the test, and informed consent is required where Article 60 provides for it, with particular protection for vulnerable persons. Data protection law applies alongside the Act and must be checked for the specific test.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689, as amended ↗AI Act Service Desk: Article 49 registration ↗AI Act Service Desk implementation timeline ↗