Instructions for use and transparency to the deployer: Article 13

Short answer: Article 13 of the AI Act requires providers to design high-risk AI to be transparent enough for the deployer to interpret the output and use it correctly, and to supply instructions for use. These address the professional user (the organisation deploying the system), not the end public. They are the link that lets the deployer meet its own obligations.

Two connected requirements

Article 13 combines two things. First, built-in transparency: the system must operate sufficiently intelligibly for a user to understand and appropriately use its output. Second, instructions for use in an intelligible format, in a language the deployer understands. Together they make human oversight possible: without knowing what a system can and cannot do, you cannot responsibly oversee it.

What the instructions must contain

The instructions must state, among other things:

• the identity and contact details of the provider;
• the intended purpose of the system;
• the level of accuracy, robustness and cybersecurity against which it was tested (see Article 15);
• known or foreseeable circumstances that may lead to risks, including reasonably foreseeable misuse;
• the performance on specific groups on which the system is intended to be used;
• the human oversight measures, including technical means to interpret the output;
• the computational and hardware resources needed, the expected lifetime and maintenance or update measures;
• where relevant, a description of the logging function (see Article 12).

Distinct from transparency to end users

Do not confuse Article 13 with the transparency duties towards natural persons (such as disclosing that someone is talking to a chatbot, or labelling AI content). Those sit elsewhere in the AI Act. Article 13 is specifically about information from provider to deployer, so that the deployer can comply with the rules and deploy the system safely.

What to do

• Write the instructions for practice: concrete, in intelligible language, not a legal annex.
• State performance and limits honestly, including foreseeable misuse and groups on which the system performs less well.
• Describe the oversight and stop measures so the deployer can implement Article 14.
• Keep the instructions current through updates and version changes.
• Connect them to your AI governance framework and to the timeline of obligations.

The instructions for use are a fixed requirement in the overview of high-risk obligations and are taken into account in the conformity assessment and CE marking. A system no one can use properly is not compliant.