Skip to main content

Deployer duties

Published and reviewed

AI Act Article 26: deployer obligations for AI agents and the FRIA

If your company runs an AI agent it did not build, the EU AI Act probably makes it a deployer. For a high-risk agent, Article 26 lists what a deployer must do, and Article 27 adds a fundamental rights impact assessment (FRIA). Both apply to Annex III systems from 2 December 2027.

This guide applies them to an agent that answers, recommends or acts for a company, often by calling tools, such as an HR screening agent that ranks job applications. It is not legal advice.

Who is a deployer

Article 3 defines both roles:

(3) ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge; (4) ‘deployer’ means a natural or legal person, public authority, agency or other body using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity; (Article 3(3) and (4))

The test is authority, not ownership. A company that runs a vendor's agent on its own customers or candidates is the deployer, and the vendor that placed the agent on the market is the provider. Under Article 111(2), a system put into service before 2 December 2027 falls under these duties only if its design changes significantly afterwards.

Article 26 applies only to high-risk systems. A customer support assistant that answers questions about orders is not on the Annex III list. An HR agent that filters and ranks job applications is, under point 4(a). The Annex III guide draws that line. The transparency duties of Article 50 apply either way, and have since 2 August 2026. The Article 50 guide covers them.

The Article 26 duties one by one

Paragraph 8 concerns public authorities and paragraph 10 law enforcement, so neither binds a private company. The others apply.

What each paragraph asks of an agent

What each paragraph asks of an agent
ParagraphWhat it asksWhat it means for an agent
26(1)Use the system as its instructions for use sayCompare the provider's instructions with the agent as configured: its tools, its data, its customers
26(2) and (3)Assign oversight to people with the competence, training, authority and supportSomeone who can read the agent's tool calls and switch it off
26(4)Keep input data you control relevant and representativeFor an agent that looks up your own documents before it answers, those documents are input data you control
26(5)Monitor, inform the provider, report a risk and suspend useAn escalation path with written suspension criteria
26(6)Keep the logs under your control for at least six monthsThe sequence of tool calls, with inputs and outputs, tied to a version of the agent
26(7)Inform workers before using a high-risk system at workNotice to staff representatives and the staff affected
26(9)Use the provider's information for the data protection impact assessmentYour data protection officer (DPO) has the instructions for use
26(11)Inform the people the system makes or helps make decisions aboutA notice to each applicant whose file the agent scores
26(12)Cooperate with the authoritiesA named contact and an evidence file ready to hand over

Two paragraphs need a closer read. Paragraph 5 sets the monitoring duty and what to do when a risk appears:

Deployers shall monitor the operation of the high-risk AI system on the basis of the instructions for use and, where relevant, inform providers in accordance with Article 72. Where deployers have reason to consider that the use of the high-risk AI system in accordance with the instructions may result in that AI system presenting a risk within the meaning of Article 79(1), they shall, without undue delay, inform the provider or distributor and the relevant market surveillance authority, and shall suspend the use of that system. (Article 26(5))

Write down in advance what counts as such a risk for your agent, who decides to suspend it, and how.

Paragraph 11 is a second notice, separate from Article 50:

Without prejudice to Article 50 of this Regulation, deployers of high-risk AI systems referred to in Annex III that make decisions or assist in making decisions related to natural persons shall inform the natural persons that they are subject to the use of the high-risk AI system. (Article 26(11))

Article 50 tells a customer they are talking to an AI. Paragraph 11 tells an applicant that a high-risk system is used on them.

When a deployer becomes a provider

Under Article 25(1), a deployer becomes the provider of a high-risk system, with the duties of Article 16, in three cases.

The three triggers

The three triggers
TriggerAgent exampleConsequence
Your name or trademark on a high-risk system already on the marketThe company sells a vendor's HR screening agent under its own brandThe company carries the provider duties, unless the contract allocates them otherwise
A substantial modification to a high-risk system that stays high-riskThe company retrains the vendor's screening agent or changes how it ranks candidatesThe company carries the provider duties, including the conformity assessment
A changed intended purpose that makes a system high-riskA general-purpose assistant gets a tool that reads CVs and recommends who to interviewThe company becomes provider of an Annex III point 4(a) system

The third case catches agents most often. As provider, you then owe the quality management system of Article 17 and the technical documentation of Annex IV. Under Article 25(2), the initial provider must cooperate and share the technical documentation, unless it clearly specified that its system is not to be turned into a high-risk one. Read the vendor's terms before you change what the agent does.

The fundamental rights impact assessment

Article 27(1) requires a fundamental rights impact assessment before first use from three groups: public bodies, private entities providing public services, and deployers of Annex III point 5(b) and 5(c) systems, which cover credit scoring and life and health insurance pricing whatever the deployer's sector. A public employer that screens applicants with an agent is in. A private company that screens its own applicants is not, unless it provides a public service.

Six elements written for an agent

Six elements written for an agent
FRIA element (Article 27(1))What to write for an agent
(a) The processes the system is used inThe steps where the agent acts, such as the first screening of applications, and the tools it can call
(b) The period and frequency of useFrom which date the agent runs, and how many decisions it supports each month
(c) The people and groups affectedApplicants and staff, including groups such as people returning to work or applicants with a disability
(d) The specific risks of harmWrong refusals, unequal treatment or a referral promised and never made, drawn from the provider's Article 13 information
(e) The human oversight measuresWho reviews the agent's recommendations, which decisions a person always takes, and how to stop it
(f) The measures if a risk materialisesSuspension, remedies for the people affected, internal governance and the complaint route

The duty attaches to first use, and the assessment must be updated when an element changes (Article 27(2)). You notify the market surveillance authority of the results (Article 27(3)) and may cross-reference the data protection impact assessment (Article 27(4)). Vidimus does not write the assessment. Its evidence pack gives you tested behaviour that the assessment can cite.

A deployer's checklist

Article 26 asks for measures, people, monitoring and logs rather than a report, but a supervisor will ask you to show what you did.

  1. Classify each agent: one that filters job applications is Annex III, point 4(a), as the Annex III guide explains.
  2. Decide your role: deployer only, or provider under Article 25(1).
  3. Obtain the instructions for use. If you cannot, record that as a finding.
  4. Assign oversight to named people with the authority to stop the agent. The Article 14 guide covers what they must be able to do.
  5. Set up the Article 26(5) monitoring, and write down the suspension path.
  6. Confirm log retention of at least six months, tool calls included.
  7. Draft the Article 26(11) notice beside the disclosure the Article 50 guide describes.
  8. If Article 27 applies to you, carry out the assessment before first use and notify the authority.
  9. Test the agent on the version you run, ten attempts per test, and repeat after any change of model, prompt, tools or the documents it reads. How we test explains why ten.

The GDPR applies to the same agent as soon as it handles personal data, and the GDPR guide says which of its duties a test can show. Our high-risk page shows how Vidimus checks each duty: behaviour on the live agent and process in your documents.

To see which of these duties apply to your agent, run the free AI Act risk classifier.

Quick answers

Are we the deployer or the provider of a vendor’s agent?

The deployer, if you use it under your authority and the vendor placed it on the market. You become its provider if you put your name on it, modify it substantially or change its purpose so it becomes high-risk (Article 25(1)).

When do the Article 26 duties apply?

From 2 December 2027 for high-risk systems listed in Annex III. A system already in service falls under them only if its design changes significantly afterwards.

Does every deployer need a fundamental rights impact assessment?

No. Public bodies, private entities providing public services and deployers of systems under Annex III, point 5(b) or 5(c), do. It comes before first use, and its results go to the market surveillance authority.

How long must we keep the logs?

At least six months for the logs under your control, unless other law says otherwise. For an agent, keep the tool calls with their inputs and outputs.

Is the fundamental rights impact assessment the same as a data protection impact assessment (DPIA)?

No. A DPIA covers the processing of personal data, while the fundamental rights impact assessment covers all the fundamental rights of the people affected. It may cross-reference the DPIA (Article 27(4)).

Sources

Vidimus tests the duties that show in your agent’s behaviour, ten attempts per test, and files the result in a signed evidence pack.

Last reviewed