Skip to main content

Explainer

EU AI Act Article 26: Deployer Obligations for High-Risk Systems, and the Start Date That Moved to 2027

By AInews Desk
EU AI Act Article 26: Deployer Obligations for High-Risk Systems, and the Start Date That Moved to 2027
5 min read
923 words

A bank that licenses a credit-scoring model did not build it. Neither did an HR team that buys CV-screening software. The AI Act Service Desk sums those obligations up this way: deployers must use these systems according to instructions, with competent human oversight, and monitor their operation.

A leased forklift is a fair comparison

A deployer is responsible for the parts of the system it actually runs, not for how the system was designed.

A warehouse that leases a forklift works the same way. It didn't engineer the machine, and it isn't liable for the hydraulics. It does have to follow the operator's manual, put qualified drivers in the seat, keep inspection records, and take the forklift out of service when something looks wrong.

The input-data rule shows the logic most clearly. The duty applies only to the extent the deployer exercises control over the input data. Where it does, the deployer must make sure that data is relevant and sufficiently representative in view of the intended purpose of the system. If an employer feeds its own applicant records into a screening tool, that data is the employer's responsibility. How the model was trained is the provider's.

Day-to-day duties

Monitoring is tied to the provider's documentation. Deployers must watch the operation of the high-risk AI system on the basis of the instructions for use and inform providers where relevant. The summary on the Service Desk page also requires deployers to keep logs for at least six months.

DutyWhat it asks of the deployer
Follow instructionsUse the system as the provider directs
Human oversightAssign competent people to oversee it
Input dataRelevant, representative data you control
MonitoringWatch operation per instructions for use
LogsKeep for at least six months
CooperationWork with authorities on regulatory actions

The last row is easy to miss. The Service Desk summary says deployers must also cooperate with authorities on regulatory actions. In practice, the logs and monitoring records are what you would hand over.

A forklift parked in an empty warehouse aisle, surrounded by orange safety cones, as if taken out of service for inspection.A forklift parked in an empty warehouse aisle, surrounded by orange safety cones, as if taken out of service for inspection.

Who has to be told, and how fast?

  • A suspected risk. If deployers have reason to think that using the system as instructed may make it present a risk, they must without undue delay, inform the provider or distributor and the relevant market surveillance authority, and shall suspend the use of that system. A suspicion is enough to stop using the system. You don't need proof.
  • A serious incident. Deployers must immediately inform first the provider, and then the importer or distributor and the relevant market surveillance authorities. The order is set: the provider comes first.
  • Workers. Before putting into service or using a high-risk AI system at the workplace, employers must inform workers' representatives and the affected workers. That must happen, where applicable, under Union and national law and practice on information of workers. It is a step that comes before going live, not something to sort out afterward.
  • Affected individuals. Deployers of Annex III systems that make decisions or assist in making decisions related to natural persons shall inform the natural persons that the system is being used on them. That covers job applicants screened by software and borrowers scored by a model.

Public authorities have one more check. They must meet the registration obligations referred to in Article 49. If a system they plan to use has not been registered in the EU database referred to in Article 71, they shall not use that system and must inform the provider or distributor. A separate paragraph sets prior-authorization rules for post-remote biometric identification in criminal investigations, including a no later than 48 hours deadline. That paragraph does not apply to most commercial deployers.

When do the obligations start?

The Regulation as a whole shall apply from 2 August 2026, and Chapters I and II have applied from 2 February 2025. But Article 26 sits in Section 3 of Chapter III, and the Digital Omnibus amendments moved that chapter's start. Article 113 now applies Chapter III, Sections 1, 2 and 3 from:

  • 2 December 2027 for AI systems classified as high-risk under Article 6(2) and Annex III. Annex III covers uses like hiring and credit scoring, so this is the date for the bank and the HR team above.
  • 2 August 2028 for AI systems classified as high-risk under Article 6(1) and Annex I.

The Service Desk marks both lines AMENDED, and says its text is based on the EUR-Lex consolidated version of the AI Act as at 27 July 2026. Before you fix an internal compliance date, check which annex your system falls under.

What to do now

You need people who understand the system and have the authority to override it. You need data you can defend, records kept for at least six months, and a clear process for suspending the system and notifying the right people in the right order. For an employer, add one fixed step before go-live: telling the workforce. None of this depends on the final date, and the processes will take time to build.

Sources

5 min read
•
923 words

Tags

EU AI Act Article 26 deployer obligations high-risk systemshigh-risk AI system requirements compliancedeployer responsibilities EU AI Actwhen does AI system deployment begin

Related Articles