The screening order is scope, role, prohibition, Annex route, transparency and general purpose model duties, and open questions, in that order, with an evidence record for each step and a confirm with authority exit for borderline questions. The evidence record should state the system, the version, the intended purpose, the facts, the provisions considered, the reasons, the evidence relied on, the reviewer, the date, and the open questions with owners and dates. Key terms in plain language, tied to Article 3 definitions, include AI system, provider, deployer, importer, distributor, authorised representative, placing on the market, putting into service, intended purpose, reasonably foreseeable misuse, safety component, substantial modification, and general purpose AI model. This guide gives the order, the record and the terms as a working tool. It does not decide the classification of any system.
What is the screening order, step by step?
The screening order is: scope, role, prohibition, Annex route, transparency and general purpose model duties, and open questions. Scope asks whether the system is an AI system under Article 3(1), whether it is placed on the Union market, put into service in the Union, or its output is used in the Union under Article 2, and whether an exclusion applies, including the research, military and open source exclusions in Article 2, subject to their limits. Role asks whether each person is a provider, a deployer, an importer, a distributor, a product manufacturer or an authorised representative, under Articles 3, 22 to 25, 28 and 54, and whether a value chain change has made a person a provider under Article 25. Prohibition asks whether Article 5 prohibits the practice, with the full Article text and the facts, and with a pause and confirmation where the screen is borderline. Annex route asks whether Article 6(1) and Annex I, or Article 6(2) and Annex III, classify the system as high risk, subject to the Article 6(3) filter and the profiling condition, with the Commission guidelines checked. Transparency and general purpose model duties ask whether Article 50 applies, and whether Chapter V applies to a general purpose model, even where the system is not high risk. Open questions ask what evidence is missing, what guidance is needed, and what authority or adviser should confirm the position, with owners and dates. That order should be followed for every system, in writing, with each step recorded, because a step that is skipped will be missed, and a step that is done out of order will waste work.
What should the evidence record contain?
The evidence record should contain, for each system: the system name, version and description, the intended purpose, the provider, deployer, importer, distributor and authorised representative, the scope analysis with the Article 2 and Article 3 provisions, the prohibition screen with the Article 5 list and reasons, the Annex analysis with the Annex I and III categories considered and rejected, the Article 6(3) filter analysis where relevant, the transparency and general purpose model analysis, the standards, codes and guidance relied on with versions and dates checked, the evidence relied on including instructions, marketing, data, tests and validation, the reviewer and date, the application dates that apply, and the open questions with owners, dates and actions. The record should be kept in a single register, with a version history, so that a reviewer can see what changed, when, and why. The record should be signed or approved by a named person, with the approval date, and should be available to an authority, a notified body, a customer or an auditor on request, subject to confidentiality. A record that is kept in several places, with different versions in each, will create confusion. A record that is kept only in the head of one person will be lost when that person leaves. The record is the evidence that the screening was done, and it should be treated with the same care as the conformity file.
For the broader regulatory framework, see questions and misconceptions overview.
When should the record be reviewed?
The record should be reviewed at a planned interval, and on defined triggers. The planned interval should be short enough to catch changes that affect the classification, and long enough to be sustainable, such as quarterly or six monthly for a small portfolio, with a shorter interval for a system that is approaching placing on the market or that is in a sandbox or test. Triggers for an immediate review include a change to the intended purpose, a substantial modification, a new function, a new data source, a new market, a Regulation amendment, a new standard, a new guideline, a serious incident, or an authority question. The review should be recorded with the date, the reviewer, the result, and the action taken, including a nil return where no change was found. The review should consider whether the prohibition, Annex, transparency and general purpose model conclusions are still correct, whether the evidence is still current, and whether the open questions have been resolved. A record that is not reviewed will become out of date without anyone noticing, and a system that was correctly classified when it was launched can become incorrectly classified after a change. The review interval and triggers should be written into the governance procedure, with a named owner, so that review does not depend on memory.
What do the key terms mean in plain language?
Key terms in plain language, tied to Article 3 definitions, include the following. An AI system is a machine based system designed to operate with varying levels of autonomy, that may exhibit adaptiveness, and that, for explicit or implicit objectives, infers from the input it receives how to generate outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments. A provider is a person who develops an AI system or a general purpose model, or has one developed, and places it on the market or puts it into service under their own name or trademark. A deployer is a person using an AI system under their own authority, except where the system is used in the course of a personal non professional activity. An importer is a person located or established in the Union that places on the market an AI system that bears the name or trademark of a person established in a third country. A distributor is a person in the supply chain, other than the provider or importer, that makes an AI system available on the Union market. An authorised representative is a person established in the Union who has received a written mandate from a provider to perform the tasks in the mandate. Placing on the market is the first making available on the Union market. Putting into service is the supply for first use directly to the deployer or for own use in the Union for the intended purpose. Intended purpose is the use for which the system is intended by the provider, including the specific context and conditions of use, as specified in the instructions, promotional material and statements. Reasonably foreseeable misuse is use not in accordance with the intended purpose, but which may result from reasonably foreseeable human behaviour or interaction with other systems. A safety component is a component that fulfils a safety function, such that its failure or malfunctioning endangers health and safety or property. A substantial modification is a change after placing on the market or putting into service which affects compliance with the Chapter III Section 2 requirements or results in a modification to the intended purpose. A general purpose AI model is an AI model that displays significant generality and is capable of competently performing a wide range of distinct tasks, regardless of how it is placed on the market, and that can be integrated into a variety of downstream systems or applications.
How should borderline questions be handled?
Borderline questions should be handled by pausing the affected use, recording the facts and the analysis, and confirming the position with a competent authority or legal adviser, with the evidence record in hand. A borderline question is one where the Article text, the guidelines and the facts do not give a clear answer, where the practice is close to a prohibition, where the Annex category is close to the facts, where the Article 6(3) filter is close to applying, or where the role or the territorial reach is unclear. The pause should be recorded, with the date, the reason, and the person who decided it. The advice should be sought in writing, with the question stated clearly, the facts given, and the analysis attached, so that the adviser can test the analysis. The advice should be kept with the record, with the action taken on it recorded. A borderline question that is resolved by a general guide, by a vendor statement, or by the practice of other providers, without authority or legal confirmation, has not been resolved. A borderline question that is resolved by a pause, a written question and a recorded answer has been resolved, even where the answer is that further evidence is needed. The confirm with authority exit is not a failure of the screening; it is the correct end of the screening for a borderline question.
Continue with misconception: all AI is high risk.
What does this tool not decide?
This tool does not decide whether any system is prohibited, high risk or subject to transparency duties, does not reproduce Article 3 or any other Article in full, and does not give a classification for any named system. The screening order, the evidence record and the key terms are a method and a structure, not a determination. A real screening needs the real facts, the consolidated text, the standards and guidance, and confirmation from a competent authority or legal adviser where the question is borderline. This guide gives the working tool, not the answer for a particular system.
Frequently asked questions
What is the screening order?
Scope, role, prohibition, Annex route, transparency and general purpose model duties, and open questions, in that order, with an evidence record for each step and a confirm with authority exit for borderline questions. The prohibition screen comes before conformity planning, because a prohibited practice cannot be made lawful by a file.
What is a substantial modification in plain language?
A change to an AI system after it is placed on the market or put into service which affects its compliance with the high risk requirements, or which changes its intended purpose. A substantial modification can make the person who makes it a provider for the modified system, and can require a new conformity assessment. The Article 3 definition and the Article 25 provisions must be checked.
What is the difference between a provider and a deployer?
A provider develops a system or model and places it on the market or puts it into service under their own name or trademark. A deployer uses a system under their own authority, except for personal non professional use. A person can be a provider for one system and a deployer for another, and a deployer can become a provider where Article 25 applies.
When should the evidence record be reviewed?
At a planned interval, such as quarterly or six monthly, and immediately on triggers including a purpose change, a substantial modification, a new function or data source, a new market, a Regulation amendment, a new standard or guideline, a serious incident, or an authority question. Each review should be recorded with date, reviewer, result and action.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689 ↗AI Act Service Desk timeline ↗