EU AI Act Risk Classification Intake Workflow

Start with a stable record that describes the system or model, the EU connection, the operator roles, and the exact classification question. Do not classify from a product nickname alone; Article 3 classification depends on intended purpose, how the system is placed on the market or put into service, who uses it, and whether a GPAI model is integrated or supplied separately.

Create one intake record per materially different intended purpose. A recruiting screen, a customer-support assistant, a credit-scoring component, and a GPAI model API can have different AI Act routes even when they share an underlying model.

The first classification gate is not risk tier; it is whether the AI Act applies to the fact pattern. Article 2 covers providers placing AI systems or GPAI models on the Union market or putting AI systems into service in the Union, deployers established or located in the Union, third-country providers and deployers where output is used in the Union, importers, distributors, product manufacturers, authorised representatives, and affected persons located in the Union.

The same screen must capture exclusions so that out-of-scope decisions are not overbroad. Article 2 exclusions include areas outside Union law, military, defence or national security purposes, certain international law-enforcement or judicial cooperation uses, AI systems or models specifically developed and put into service solely for scientific research and development, pre-market research, testing or development activity other than real-world testing, purely personal non-professional deployer use, and free/open-source AI systems unless they are placed on the market or put into service as high-risk systems or fall under Article 5 or Article 50.

Run the Article 5 screen before high-risk classification. A prohibited-practice hit should stop normal intake and move to legal escalation, product withdrawal, redesign, or non-use review rather than being treated as a manageable high-risk obligation.

Capture the exact practice, deployment context, affected persons, purpose, expected effect, data used, safeguards claimed, and whether any narrow law-enforcement biometric exception is being asserted.

If no Article 5 prohibition is identified, classify high-risk status through both Article 6 routes. Article 6(1) covers AI used as a safety component of a product, or as a product itself, where the product is covered by Annex I Union harmonisation legislation and requires third-party conformity assessment. Amended Article 6 excludes systems used solely for non-safety user assistance, performance optimisation, service efficiency, automation, convenience, or quality control from the safety-component route, while confirming that a system whose failure or malfunction would endanger health and safety is a safety component. A product does not meet Article 6(1)(b) where third-party assessment is required only for risks unrelated to health and safety. Article 6(2) covers AI systems listed in Annex III.

For Annex III, intake must capture the use-case area and the Article 6(3) derogation analysis. A system listed in Annex III can be documented as not high-risk only where it does not pose a significant risk of harm to health, safety, or fundamental rights, including by not materially influencing decision-making, and one of the Article 6(3) conditions applies. Profiling of natural persons remains high-risk.

Run Article 50 even when the system is neither prohibited nor high-risk. The duties attach to particular functions: direct interaction with people, generation of synthetic audio, image, video, or text, emotion recognition or biometric categorisation, and deployment of deepfakes or certain AI-generated text published to inform the public on matters of public interest.

Assign the duty to the correct actor. Providers must design covered interactive or generative systems so the required disclosure or machine-readable marking can be delivered. Deployers carry the disclosure duties for emotion recognition, biometric categorisation, deepfakes, and covered public-interest text. The disclosure must be clear and distinguishable no later than the first interaction or exposure, subject to the article's law-enforcement, artistic, satirical, and editorial exceptions.

Run a separate GPAI screen whenever the intake involves a model supplied for integration, an API model, a foundation model, a modified model, or an AI system based on a GPAI model. Article 3 distinguishes a general-purpose AI model from a general-purpose AI system and defines downstream providers that integrate AI models into AI systems.

The GPAI screen should not replace system classification. A downstream AI system built on a GPAI model can still need Article 5, Article 6, Annex III, Article 50, deployer, and registration analysis.

Close the intake by assigning role-specific owners and reopening conditions. Article 25 can shift provider responsibilities to a distributor, importer, deployer, or other third party that puts its name or trademark on a high-risk AI system, substantially modifies a high-risk system, or changes the intended purpose of an AI system so that it becomes high-risk.

Treat classification as a maintained record. Reassess when the intended purpose, EU market path, user population, model supplier, product integration, Annex III use case, profiling status, biometric function, GPAI capabilities, training compute, documentation, or post-market evidence changes.

Record legal status with every application date. The original Article 5 prohibitions have applied since 2 February 2025, the content-related Article 5 additions apply from 2 December 2026, GPAI provider obligations have applied since 2 August 2025, and Article 50 applies from 2 August 2026. Regulation (EU) 2026/1744 entered into force on 27 July 2026 and applies Chapter III Sections 1-3 from 2 December 2027 for Article 6(2) Annex III systems and from 2 August 2028 for Article 6(1) Annex I systems.