Real-World Testing of High-Risk AI: Article 60 and the Safeguards That Apply
Real-world testing of high-risk AI under Article 60: a testing plan, registration, a 6-month cap, oversight, and Article 61 informed consent.
The EU AI Act, Regulation (EU) 2024/1689, lets you test a high-risk AI system in real-world conditions before it goes to market — outside a regulatory sandbox — under Article 60. This is not a loophole for skipping conformity assessment. It is a supervised route to validate a system in the field, hedged with safeguards: a testing plan approved by the market-surveillance authority, registration, a hard time limit, effective human oversight, and the informed consent of test subjects under Article 61, who can withdraw at any time.
The short version: Article 60 lets you test in the wild, but on the regulator's terms.
Article 60 is not the Article 57-58 sandbox
First, a distinction people routinely blur. A regulatory sandbox (Articles 57-58) is a controlled environment a national authority runs, where you develop and test under direct supervision and SMEs get priority, free access. Article 60 real-world testing is different: you test your high-risk system in actual operating conditions, outside any sandbox, before placing it on the market — under your own conduct but with the market-surveillance authority's approval and the Act's safeguards.
| Regulatory sandbox (Art 57-58) | Real-world testing (Art 60) | |
|---|---|---|
| Setting | Controlled environment run by the authority | Actual operating conditions, outside a sandbox |
| Who runs it | National competent authority supervises | Provider/prospective provider conducts |
| SME access | Priority + free (Art 58) | Open to providers of Annex III high-risk systems |
| Approval | Admission to the sandbox | Testing plan approved by market-surveillance authority |
| Subject consent | Per sandbox rules | Informed consent required (Art 61) |
Both are innovation supports; they are not interchangeable, and citing the wrong one is a common error.
What Article 60 requires
Real-world testing is available to providers (and prospective providers) of Annex III high-risk systems, before placing on the market or putting into service. The conditions, in plain terms:
- A real-world testing plan, submitted to the market-surveillance authority in the member state where testing happens. If the authority does not respond within 30 days, the plan is deemed approved (unless national law requires explicit authorisation).
- Registration of the test in the EU database with a Union-wide unique identification number and the information in Annex IX (with a secure non-public route for law-enforcement, migration and border-control cases).
- A time limit: testing lasts no longer than necessary and in any case not more than six months, extendable once by a further six months on notice to the authority.
- Effective oversight by suitably qualified people, with the authority to perform their tasks, and the system's predictions, recommendations or decisions must be reversible and capable of being disregarded.
- Protection of vulnerable subjects, EU establishment or an EU legal representative, and safeguards on any transfer of collected data to third countries.
- Serious-incident reporting under Article 73, with suspension of the test until mitigation; market-surveillance authorities supervise the testing.
The test does not suspend the Article 5 prohibitions — you cannot field-test a prohibited practice.
Article 61: informed consent and the right to withdraw
This is the safeguard that protects the people involved. Under Article 61, subjects of real-world testing must give informed consent before participating — freely given, with clear information about the test's nature, objectives, and what participation involves. And under Article 60(5), any subject (or their legal representative) may withdraw at any time, without any detriment and without giving a reason, by revoking consent — and may request the immediate, permanent deletion of their personal data. Withdrawal does not undo activities already lawfully carried out. There is a narrow, conditioned exception for certain law-enforcement testing where seeking consent would defeat the test, hedged with strict protections.
Consent here is the field-testing analogue of medical-trial ethics: real people, real conditions, a genuine opt-out.
The deadline
The innovation-support provisions, including Article 60, carry the 2 August 2026 application date. That date is not affected by the Digital Omnibus deferral — the Omnibus, adopted in June 2026, moves the stand-alone high-risk Annex III obligations to 2 December 2027 (and product-embedded Annex I systems to 2 August 2028), but the testing framework itself runs to its own 2 August 2026 date. In practice, you would use Article 60 to validate a system ahead of whichever high-risk compliance date ultimately binds it.
Worked example: an SME validates a high-risk system
Verza Diagnostics OÜ, a 16-person Estonian start-up with roughly EUR 2M turnover, has built a triage-support tool it expects to classify as high-risk under Annex III. Before full conformity assessment, it wants evidence the tool performs in a live clinic rather than only on historical data. It uses Article 60: it draws up a real-world testing plan and submits it to the national market-surveillance authority; hearing nothing within 30 days, it proceeds on deemed approval; it registers the test under Annex IX; it caps the test at four months; it staffs clinicians as the qualified overseers who can disregard any output; and it obtains Article 61 informed consent from every participating patient, each of whom can withdraw and have their data deleted at any time. When a near-miss occurs, Verza reports it as a serious incident under Article 73 and pauses until it is resolved. By conformity assessment, the hard questions have been answered under supervision — not discovered at launch.
How Confir helps
Confir maps where a system sits against the Act and which obligations a real-world test would help you de-risk — the Article 9 risk-management and Article 10 data-governance evidence an authority will expect to see in your testing plan — so you arrive with a structured record rather than a blank page. The engine is deterministic and rule-based, and it surfaces the related SME supports (the Articles 57-58 sandbox, proportionate fees under Article 62) where they apply to you.
Frequently asked questions
What is real-world testing under the EU AI Act?
It is a route under Article 60 to test a high-risk AI system in actual operating conditions, outside a regulatory sandbox, before placing it on the market. It is available to providers of Annex III high-risk systems under safeguards: an approved testing plan, registration, a time limit, effective oversight, and the informed consent of subjects under Article 61. It is supervised validation, not an exemption from conformity assessment.
How is Article 60 different from a regulatory sandbox?
A sandbox (Articles 57-58) is a controlled environment a national authority runs, with priority, free access for SMEs. Article 60 real-world testing happens in actual operating conditions outside a sandbox, conducted by the provider but approved and supervised by the market-surveillance authority. Both support innovation, but they are distinct mechanisms — citing the sandbox articles for real-world testing is a common error.
Do test subjects have to consent to real-world testing?
Yes. Under Article 61, subjects must give informed consent before participating — freely given, with clear information about the test. Under Article 60(5) they may withdraw at any time, without detriment or justification, by revoking consent, and may request deletion of their personal data. A narrow, conditioned exception exists for certain law-enforcement testing where consent would defeat the purpose.
How long can real-world testing last?
No longer than necessary, and in any case not more than six months, extendable once by a further six months on prior notice to the market-surveillance authority with an explanation. The test must also be registered (Annex IX), overseen by qualified people who can disregard the system's outputs, and suspended if a serious incident occurs, which must be reported under Article 73.
Does the testing plan need approval?
It must be submitted to the market-surveillance authority in the member state where testing takes place. If the authority does not respond within 30 days, the plan is deemed approved — unless national law does not provide for tacit approval, in which case explicit authorisation is required. The provider must also be established in the EU or have an EU legal representative.
When does Article 60 apply?
The innovation-support provisions, including Article 60, carry the 2 August 2026 application date, which the Digital Omnibus deferral does not move. The high-risk obligations a tested system must ultimately meet apply from 2 December 2027 for stand-alone Annex III systems, and from 2 August 2028 for product-embedded Annex I systems — the dates set by the Digital Omnibus, adopted in June 2026.
Related guides
- EU AI Act regulatory sandboxes
- Conformity assessment, defined
- Annex III high-risk use cases
- Article 73: reporting serious incidents
- EU AI Act for startups
- Is the EU AI Act delayed?
Manage your EU AI Act compliance in one place
Confir automates risk classification, technical documentation, and audit trails for any company. No consultants. No 6-month projects. 14-day free trial.
Start free trial →