What are the EU AI Act risk categories?
Direct answer
The EU AI Act has four risk levels: unacceptable risk, which is banned; high risk, which carries strict obligations; transparency risk, which carries disclosure duties; and minimal or no risk, which the Act leaves unregulated. General-purpose AI models follow a separate set of rules, not a fifth level, and one system can carry more than one label.
The EU AI Act sorts AI systems into four risk levels, as the European Commission presents them: unacceptable risk, which is banned; high risk, which carries strict obligations; transparency risk, which carries disclosure duties; and minimal or no risk, for which the Act adds no rules. General-purpose AI models follow a separate set of rules, not a fifth level. One system can carry more than one label: a high-risk system that also talks to people must meet both sets of duties (Article 50(6)). Unacceptable risk covers the practices prohibited by Article 5, such as social scoring, harmful manipulation and emotion recognition at work or in education. As of 19 November 2026, the first eight prohibitions have applied since 2 February 2025; a ninth, added by the AI Omnibus, applies from December 2026. High risk covers AI used as a safety component of products under the EU laws listed in Annex I, and AI used in the areas listed in Annex III, such as recruitment, education, essential services and law enforcement. The duties apply from 2 December 2027 for Annex III and from 2 August 2028 for Annex I. Transparency risk covers chatbots, AI-generated content and deepfakes under Article 50, whose duties have applied since 2 August 2026. Minimal or no risk is everything else.
What sits in each risk level, and what does it owe?
- Unacceptable risk (Article 5). Banned: harmful manipulation and deception; exploiting vulnerabilities due to age, disability or social or economic situation; social scoring; predicting a person’s risk of committing a crime based solely on profiling; untargeted scraping of facial images; emotion recognition at work and in education; biometric categorisation to infer protected characteristics; real-time remote biometric identification by law enforcement in publicly accessible spaces, with narrow exceptions; and, from December 2026, AI that generates non-consensual intimate imagery or child sexual abuse material. The duty: do not place these on the market, put them into service or use them. Fines reach EUR 35 million or 7% of worldwide annual turnover.
- High risk (Article 6, Annexes I and III). For example CV-sorting software for recruitment, credit scoring, AI in critical infrastructure or a safety component of a medical device. Providers run risk management, data governance, technical documentation, logging, human oversight and a conformity assessment before the system goes on the market. Deployers use it as instructed, assign human oversight and keep its logs.
- Transparency risk (Article 50). For example a customer service chatbot, a generator of synthetic images or text, or a deepfake. People must be told they are dealing with AI unless it is obvious, AI-generated content must be marked in a machine-readable way, and deepfakes must be disclosed.
- Minimal or no risk. Everything else. The Act adds no rules for these systems, though the organisations that provide or deploy them still take measures to support AI literacy (Article 4).
How does a system become high-risk?
There are two routes. Under Article 6(1), an AI system is high-risk when it is a safety component of a product, or is itself a product, covered by the EU laws in Annex I, such as machinery, toys, lifts and medical devices, and that product must undergo a third-party conformity assessment. Under Article 6(2), an AI system is high-risk when it is used in one of the eight areas of Annex III: biometrics; critical infrastructure; education and vocational training; employment and workers management; access to essential private and public services; law enforcement; migration, asylum and border control; and the administration of justice and democratic processes.
Article 6(3) lets a provider conclude that an Annex III system is not high-risk when it does not pose a significant risk of harm, for example because it performs a narrow procedural task. The exception never applies when the system profiles natural persons, and the provider must document its assessment before placing the system on the market (Article 6(4)). As of 19 November 2026, the Commission’s guidelines on high-risk classification are still a draft, published on 19 May 2026. For the full test, see high-risk AI systems and the classification test.
Where do general-purpose AI models fit?
General-purpose AI models, the large models that can be built into many different systems, are regulated as models in Articles 51 to 55, apart from the four levels that apply to AI systems. Since 2 August 2025, every provider of such a model must keep technical documentation, give information to the providers that build on it, put in place a copyright policy and publish a summary of the content used for training (Article 53). Models with systemic risk, presumed when their training used more than 10^25 floating-point operations (Article 51(2)), carry extra duties: model evaluation, adversarial testing, assessing and mitigating systemic risks, reporting serious incidents and cybersecurity (Article 55).
A system built on such a model is classified by its own use. A chatbot built on a general-purpose model is a transparency-risk system; the same model used to sort job applications sits inside a high-risk system. To see what each level means for your organisation, by role and date, use the EU AI Act compliance checklist, the EU AI Act guide and EU AI Act compliance software.
Change log: 19 November 2026, first published.
How Acuna GRC handles it
In Acuna GRC, AI Governance records the risk class of every AI system, with the reasoning, in one register.
- Solution
EU AI Act
Run the EU AI Act in one register: every AI system with your role, its risk class, impact assessment, human oversight and incidents, next to ISO 42001.
- Module
AI Governance
A register of the AI systems you build, buy or deploy, with impact assessments, human oversight, data provenance, incidents and a public contestation channel.
- Solution
ISO 42001
Run ISO 42001 next to ISO 27001: register every AI system, gate approval on impact assessments, name overseers and keep the records an audit looks at.
Sources
- Regulation (EU) 2024/1689 (AI Act), Articles 4, 5, 6, 25, 50, 51, 53, 55 and 99, Annexes I and III
- Regulation (EU) 2026/1744 (Digital Omnibus on AI)
- European Commission, AI Act (updated 19 October 2026)
- European Commission, draft guidelines on the classification of high-risk AI systems (19 May 2026)
This page explains the rules in general terms and is not legal advice.
Frequently asked
Is a chatbot high-risk under the EU AI Act?
Not by itself. A chatbot is a transparency-risk system: people must be told they are interacting with AI unless that is obvious (Article 50(1)). It becomes high-risk only when it is used for an Annex III purpose, such as assessing whether a person is eligible for public assistance benefits, and then both sets of duties apply.
Who decides which risk category a system belongs to?
The provider classifies it and, for an Annex III system it considers not high-risk, documents that assessment (Article 6(4)). Deployers should check the classification against their own use: under Article 25, a deployer that substantially modifies a high-risk system, or changes the intended purpose of a system so that it becomes high-risk, becomes its provider.
See how Acuna GRC runs this.
Tell us your frameworks and your team. We show you how Acuna GRC runs the program.
Hosted in Switzerland, the EU or the US
.png&w=3840&q=75&dpl=dpl_96PxuEbyzwoWxjweCh9wLaGaDCHP)

