Skip to content
← Back to blog
AI ActArticle

AI Act Article 5: review the proposed feature before buying

A system's purpose and behaviour matter more than its product label. Prepare the feature description and unresolved questions for a prohibited-practices review.

Author

Syntalith

Published Updated 2 min read

Article 5 prohibits specified AI practices. Before buying a feature that evaluates or influences people, describe what it actually does and have that use assessed. A product name or a supplier's general compliance statement is too broad to settle the question.

This guide helps a project owner prepare for that review. It does not provide a legal opinion about a product or an exhaustive list of prohibited uses. The official sources were checked on 11 October 2026.

Read the conditions of the relevant prohibition

Article 5 addresses practices including certain harmful manipulation, exploitation of vulnerabilities, social scoring, untargeted collection of facial images for recognition databases, and particular biometric uses. The conditions and exceptions differ. For example, the workplace and education emotion-inference provision has a medical-or-safety exception. The amended text adds further prohibitions concerning certain sexual content, with their own conditions. Consult the current Article 5 text rather than applying a short category label to every system.

An organisation reviewing a feature needs to identify the applicable provision before discussing a claimed exception. A technical team can describe the inputs and behaviour; a qualified reviewer assesses their legal significance. Keep the review tied to a product version and intended use.

A fictional meeting product changes its purpose

Imagine a company buying a tool to prepare meeting notes. During procurement, the supplier proposes an additional dashboard that infers employees' emotions from their voices and presents a score to managers.

The new feature needs its own description and legal review. Calling the whole package “meeting productivity” does not explain the score. The buyer should ask what inputs it uses, what it claims to infer and which management decisions it is intended to support. This example identifies questions for assessment; it does not determine the legal status of a real product.

The team may decide it only needs a draft record of agreed actions. That narrower purchasing requirement should be reflected in the selected features and configuration. Someone should verify that the unwanted dashboard is not enabled through a default setting or a later upgrade.

Review the feature that will actually run

A procurement record is more useful when it includes a sample output and a description of who receives it. In the fictional example, a private note for the participant and a recurring employee score shown to a manager serve different purposes. The record should preserve that distinction.

Ask the supplier to identify optional modules, the settings that affect them and planned changes. If the answer remains unclear, record the uncertainty rather than interpreting it as confirmation that the feature is acceptable. The person responsible for the purchase should know which question is unresolved and who is assessing it.

Technical restrictions also need verification. A statement that a feature is disabled should be checked against the purchased configuration and the permissions available to users. That check establishes a configuration fact; it does not replace the legal assessment.

A focused preparation scope

Syntalith can help assemble the technical material for this review: the proposed workflow, feature list, sample outputs and relevant configuration questions. The agreed scope can include checking whether the selected deployment matches the restricted use described to the adviser. Legal interpretation and approval need a separately assigned specialist.

For the fictional meeting product, a useful first result would be a comparison of the note-taking requirement with the supplier's additional analytics features. The buyer could then decide what to commission after the review, with the unresolved issues visible. We do not offer this comparison as a compliance certificate.

Describe the feature being considered, without uploading recordings or employee records. Current service pricing provides the starting point for the commercial discussion. The provider and deployer guide covers the related question of responsibilities between organisations.

OpenAI Select Partner

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 work

Reduce 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
Discuss an automation