Skip to main content
Vidimus
ENFR

Free tool

EU AI Act risk classifier for AI agents

Answer a few questions about your AI agent to see its likely risk class under the EU AI Act and the main obligations that follow.

About your agent

Which of these uses is closest to what the agent does?

Annex III lists the uses that make an AI system high-risk. Each comes with an example and, where it matters, what it does not cover.

Does it talk directly with people, customers or staff (chatbot, voice assistant, agent)?

Staff are people too: an internal helpdesk chatbot also talks directly with people.

Does it generate text, images, audio or video?

Article 50(2): AI-generated outputs must be marked so machines can detect them.

Does it do anything Article 5 prohibits?

Article 5 prohibits social scoring, and manipulating people or exploiting vulnerabilities due to their age, disability or social or economic situation in a manner that causes or is reasonably likely to cause significant harm. It also prohibits inferring emotions in the workplace and in education institutions, inferring race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation from biometric data, and building facial recognition databases by untargeted scraping. Emotion recognition for medical or safety reasons is not prohibited. From 2 December 2026, generating or manipulating realistic images, video or audio of an identifiable person’s intimate parts, or of that person engaged in sexually explicit activities, without their consent, or child sexual abuse material within the meaning of Directive 2011/93/EU, is prohibited too.

Is it a safety component of a regulated product, such as a toy, a lift or a medical device?

Products covered by the EU product-safety laws listed in Annex I. Rare for an agent that serves customers or staff.

0 of 5 answered

How the EU AI Act classifies an AI system

The Act sorts AI systems by what they are used for, not by how they are built. A few practices are prohibited outright under Article 5, such as social scoring and manipulative techniques that cause or are reasonably likely to cause significant harm. A system is high-risk in two cases. It is intended to be used as a safety component of a product, or is itself a product, covered by the Union harmonisation legislation listed in Annex I, such as toys, lifts or medical devices, and that product is required to undergo a third-party conformity assessment. Or it is used for one of the purposes listed in Annex III, unless the Article 6(3) exception applies.

Systems that interact with people, generate content, recognise emotions or sort people by biometric traits carry the Article 50 transparency duties whatever their class. Everything else carries no obligation tied to its risk level, what is commonly called minimal risk: AI literacy under Article 4 still applies, and codes of conduct are voluntary. General-purpose AI models follow a separate track of provider obligations.

For a second reading, the European Commission’s AI Act Service Desk publishes its own checker: EU AI Act Compliance Checker

The eight Annex III categories, each with an example

Each category is a point of Annex III, with an example of a use that falls under it and, where it helps, one that does not.

  1. 1
    BiometricsRecognising customers’ emotions in a call centre is high-risk, while inferring emotions in the workplace, your own staff’s for example, is prohibited by Article 5 unless for medical or safety reasons. A face match that only confirms a customer is who they claim to be at onboarding is excluded.
  2. 2
    Critical infrastructureSafety components for energy, water, transport and digital infrastructure. This rarely concerns an agent that serves customers or staff.
  3. 3
    Education and vocational trainingAdmissions and exam grading at a school, university or vocational training provider. Grading your own staff falls under employment.
  4. 4
    EmploymentScreening job applicants, or an AI that allocates work to staff or rates their performance.
  5. 5
    Essential servicesCreditworthiness assessment and credit scoring (point 5(b)), and risk assessment and pricing in life and health insurance (point 5(c)).
  6. 6
    Law enforcementTools used by or for the police and prosecutors. A company’s own fraud detection is not law enforcement.
  7. 7
    Migration, asylum and border controlTools used by or on behalf of migration, asylum or border authorities.
  8. 8
    Justice and democratic processesTools used by courts or to influence an election.

What “high-risk” means for a deployer before December 2027

The Digital Omnibus, Regulation (EU) 2026/1744, which amended the AI Act in 2026, deferred the high-risk obligations for Annex III systems to 2 December 2027, and to 2 August 2028 for AI built into Annex I products. Until then a deployer has no high-risk duty to meet, but regulators and auditors already ask for documented AI governance. That documentation cannot be rebuilt after the fact.

From December 2027 a deployer must use the system as instructed, assign human oversight to competent people, keep the logs, and inform affected workers and the people the system makes or helps make decisions about. Deployers that assess creditworthiness or price life or health insurance, and deployers that are bodies governed by public law or private entities providing public services, must also complete a fundamental rights impact assessment under Article 27 before first use.

Article 50 applies now, and so does the GDPR whenever the agent handles personal data.

Know the class and then test the agent

A classification says which duties apply. A pilot shows whether your live agent meets them and ends with a signed evidence pack.

Request a two-week pilotCheck every Article 50 duty

Pilot

  • One agent from end to end
  • Ends with a walkthrough of the evidence pack with your risk and compliance leads
  • Credited against the first year