AI Act Article 6 and Annex III: describe the use before classifying it
A department name does not settle whether an AI system is high-risk. Prepare the intended use, output and decision process for a classification review.
Syntalith
Article 6 and Annex III work together when assessing whether an AI system falls into a high-risk category. Start with the system's intended use and the way its output affects a decision. A label such as “HR assistant” leaves too much unspecified for a useful assessment.
For a buyer, the immediate task is to describe the application accurately enough for a qualified reviewer. This guide explains that preparation. It does not classify your proposed system. Sources were checked on 11 October 2026.
Two routes into the assessment
Article 6(1) concerns certain AI products or safety components associated with the legislation in Annex I and a required third-party conformity assessment. Article 6(2) refers to the uses listed in Annex III. The amended text contains further conditions concerning safety components. Read the conditions together in Article 6.
Annex III includes specified uses in areas such as employment, education, access to essential services and justice. For employment, its examples include analysing and filtering applications and evaluating candidates. These are descriptions of purposes, rather than a rule that every application used by an HR department has the same status. See Annex III.
The legal reviewer also needs to consider prohibited practices and other applicable rules. A classification worksheet should leave room for unresolved questions instead of forcing every project into a convenient category.
One fictional assistant, two different jobs
Imagine a company evaluating software for its recruitment team. The initial description says that it helps staff find the current interview-room booking instructions. A later proposal adds a ranked list of applicants, with the highest-scoring people shown first.
Those proposals produce different outputs for different decisions. An assessment of the first description should not silently become the assessment of the ranking feature. The project owner needs to explain what the score measures, which information feeds it and how recruiters see and use the result.
We would prepare the two scopes side by side. The useful material includes a sample output, the intended users, the people affected and the point at which a person makes a decision. Screens showing the actual order of information can reveal more than a broad claim that a human remains involved.
An exception needs a reasoned assessment
Article 6(3) provides conditions under which an Annex III system may not be treated as high-risk. The overarching risk condition matters alongside the listed tasks. An Annex III system that profiles natural persons is always treated as high-risk under that provision. Article 6(4) requires the provider to document an assessment when considering an Annex III system non-high-risk. A “preparatory task” label alone does not settle the issue. Article 6
For the fictional ranking feature, the team should record what happens when a reviewer disagrees with the output and whether lower-ranked applications still receive consideration. These are facts for the assessment. Adding an approval button does not answer those questions by itself.
What to include in the implementation enquiry
Syntalith can scope the technical description, sample workflow and change record that support a classification review. Your team identifies the intended task and decision owner. A qualified adviser assesses the legal category and applicable obligations; that advice is a separate responsibility.
A proposed first engagement could compare the current workflow with the feature being considered and identify missing supplier information. It should state whether the work ends with that material or includes changes after the review. We do not present this preparation as a conformity assessment or certification.
Bring one proposed use and an example of the decision it supports. There is no need to supply candidate records in the first enquiry. See service pricing or describe the project. The provider and deployer guide covers the separate question of which organisation has which role.
Syntalith is an OpenAI Select Partner in the OpenAI Partner Network.
We help your team respond to customers faster and find information in company documents. We choose and set up the right AI tools, then teach your team how to use them.
How to introduce ChatGPT and OpenAI at workReduce manual work in a defined process
Start with where work gets stuck and who has to repair it. We will compare the available improvements with implementation and operating costs. Automating one process starts at €1,750 net, and the first call is free.
Explore AI automation