Glossary · EU AI Act: high-risk AI
Instructions for use (AI Act)
Also known as: instructions for use, IFU, user information
Under the EU AI Act, instructions for use are the information a provider must supply with a high-risk AI system so that deployers can interpret its output and use it appropriately. They must be concise, complete, correct and clear, and include the system’s intended purpose, accuracy, limitations, human oversight measures and maintenance.
- AI regulation
- EU
- Technical documentation
In one sentence
AI Act instructions for use (Art. 13) tell deployers what a high-risk AI system does, how accurate it is, where it fails and how to oversee it.
Example
The instructions for an AI triage system state its sensitivity and specificity, that it was not validated for children under 12, and which cases a clinician must always review.
What the instructions must contain
- Identity and contact details of the provider.
- Characteristics, capabilities and limitations of performance: intended purpose, level of accuracy with its metrics, robustness and cybersecurity, known or foreseeable circumstances that may lead to risks, capabilities to explain output, performance for specific persons or groups, and specifications for input data.
- Predetermined changes to the system and its performance.
- Human oversight measures, including technical measures that help deployers interpret the output.
- Computational and hardware resources, expected lifetime, and maintenance and care measures, including software updates.
- Where relevant, how deployers can collect, store and interpret the logs.
How it applies
- Technical documentation: This is the article of the AI Act closest to classic technical communication. Instructions must be “concise, complete, correct and clear” and “relevant, accessible and comprehensible” to deployers — requirements that apply the principles of standards such as IEC/IEEE 82079-1 to AI.
- Accuracy and limitations: Unlike ordinary manuals, the instructions must state performance in numbers and describe where the system does not work well. That requires close cooperation between writers and data scientists.
- Format: Instructions may be provided in digital form, which suits structured, versioned content delivered with the product.
Compared with the USA, Canada and China
The USA has no federal requirement for instructions for AI systems; the NIST AI RMF recommends documenting limitations and intended use as good practice. Canada's voluntary code asks developers to publish information on capabilities and limitations. China's generative AI measures require providers to state the intended users, occasions and purposes of a service and to guide users to use it lawfully.