1. Who it applies to
Regulation (EU) 2024/1689 applies along the whole AI value chain:
- Providers: organisations that develop an AI system or model and place it on the EU market or put it into service under their own name.
- Deployers: organisations or individuals that use an AI system in a professional capacity, such as a clinic using a triage tool or a firm using AI in recruitment. People affected by the system are not deployers.
- Importers and distributors that bring AI systems to the EU market.
It also reaches outside the EU: providers and deployers in other countries are covered when the output of their AI system is used in the EU.
2. The four risk levels
3. Banned practices (Article 5)
These uses have been prohibited since 2 February 2025, except the last item on this list, which applies from 2 December 2026:
- Subliminal, manipulative or deceptive techniques that distort decisions and cause, or are likely to cause, significant harm.
- Exploiting vulnerabilities linked to age, disability or social or economic situation.
- Social scoring that leads to unjustified or disproportionate detrimental treatment.
- Predicting the risk that a person will commit a crime based solely on profiling or personality traits.
- Building facial recognition databases by untargeted scraping of images from the internet or CCTV.
- Inferring emotions in the workplace or in education, except for medical or safety reasons.
- Biometric categorisation that infers sensitive characteristics such as race, political opinions, religion or sexual orientation, with narrow exceptions.
- Real-time remote biometric identification in public spaces by law enforcement, except in strictly listed cases with prior authorisation.
- Generating non-consensual intimate imagery and child sexual abuse material (from 2 December 2026).
4. High-risk AI (Article 6)
An AI system is high-risk in two situations:
- Product safety (Annex I): the AI is a product, or a safety component of a product, covered by the EU product laws in Annex I, and that product needs a third-party conformity assessment. This includes medical devices and in vitro diagnostic devices under the MDR and IVDR.
- Listed use cases (Annex III): biometrics; critical infrastructure; education; employment and management of workers; access to essential public and private services; law enforcement; migration and border control; and the administration of justice and democratic processes.
Annex III expressly includes several health-related uses: deciding eligibility for public benefits and services including healthcare, emergency patient triage and the dispatch of emergency services, and risk assessment and pricing in health and life insurance.
An Annex III system is not high-risk if it only performs a narrow procedural or preparatory task or improves the result of work already done by a person, unless it profiles individuals. A provider that relies on this exception must document its assessment before placing the system on the market.
5. Duties of providers of high-risk AI
- A risk management system covering the whole lifecycle.
- Data governance: training, validation and testing data that is relevant, representative and as free of errors as possible.
- Technical documentation, kept up to date.
- Automatic logging of events.
- Clear instructions for use for deployers.
- Design for effective human oversight.
- Appropriate accuracy, robustness and cybersecurity.
- A quality management system.
- Conformity assessment, an EU declaration of conformity and CE marking before the system is placed on the market.
- Registration in the EU database, and corrective action and reporting of serious incidents after launch.
6. Duties of deployers of high-risk AI (Article 26)
This is the part that matters most for clinics, insurers, employers and professional firms that buy AI rather than build it:
- Use the system according to the provider's instructions for use.
- Assign human oversight to people with the necessary competence, training, authority and support.
- Where you control the input data, make sure it is relevant and sufficiently representative.
- Monitor the system. If it may present a risk, inform the provider and the market surveillance authority and suspend its use; report serious incidents.
- Keep the logs the system generates for at least six months, unless other law requires a different period.
- Inform workers and their representatives before using high-risk AI at work.
- Use the provider's information to carry out a data protection impact assessment under the GDPR where one is required.
- Tell people when a high-risk system listed in Annex III is used to make, or help make, decisions about them.
- Public bodies must also register their use in the EU database.
7. AI literacy (Article 4)
Since 2 February 2025, every provider and deployer, whatever the risk level, must take measures to ensure a sufficient level of AI literacy among staff and others who operate or use AI systems on its behalf. The measures should reflect their knowledge, experience and training, the context of use and the people affected. No specific level is guaranteed for any individual, but you should be able to show what you have done, for example training and internal guidance.
8. Transparency (Article 50)
From 2 August 2026:
- People must be told when they are interacting with an AI system such as a chatbot or virtual assistant, unless this is obvious.
- Providers of generative AI must mark synthetic audio, images, video and text in a machine-readable way. Systems already on the market before 2 August 2026 have until 2 December 2026.
- Deployers of emotion recognition or biometric categorisation must inform the people exposed to it.
- Deepfakes, and AI-generated text published to inform the public on matters of public interest, must be disclosed, with lighter rules for artistic works and text that has been reviewed by a person who holds editorial responsibility.
Disclosure must be clear and given at the first interaction. A hidden footer or faint label is not enough.
9. General-purpose AI models
Providers of general-purpose AI models (such as large language models) have had obligations since 2 August 2025: technical documentation, information for downstream providers, a copyright policy and a public summary of training content. Models placed on the market before that date have until 2 August 2027.
Models trained with more than 1025 floating-point operations are presumed to present systemic risk and must also be evaluated (including adversarial testing), have their systemic risks assessed and mitigated, report serious incidents and be adequately protected against cyberattacks.
The voluntary General-Purpose AI Code of Practice (published 10 July 2025) helps providers show compliance. It has chapters on transparency, copyright and, for systemic-risk models only, safety and security.
10. AI in healthcare
- Medical devices: software with a medical purpose, such as diagnosing, monitoring or managing a condition, can be a medical device under the MDR. If it needs a third-party conformity assessment, the AI is high-risk under the AI Act, and both laws apply. Software intended only for lifestyle and well-being is not a medical device.
- Triage and access to care: AI used for emergency patient triage or to decide eligibility for healthcare is high-risk (Annex III).
- Insurance: AI used for risk assessment and pricing in health and life insurance is high-risk.
- Mental health and therapy AI: depending on its design and claims, it may be banned (if it exploits vulnerabilities and causes significant harm), high-risk (if it is a medical device) or subject to transparency duties. Users must know they are talking to AI, and repeated reminders may be needed for people in distress.
- Emotion recognition in workplaces and education is banned, but there is an exception for medical and safety reasons.
- Health data: the GDPR, the EHDS and, in Cyprus, the Electronic Health Law continue to apply alongside the AI Act.
11. Small and medium-sized businesses
- Priority, free access to AI regulatory sandboxes, where new AI can be tested under the supervision of an authority. Firms that follow the sandbox plan in good faith are not fined for infringements during testing.
- Conformity assessment fees proportionate to company size, and simplified technical documentation forms.
- Fines capped at the lower of the fixed amount and the percentage of turnover, instead of the higher.
- Dedicated guidance channels and training in each Member State.
12. Who enforces it
- The AI Office in the European Commission supervises general-purpose AI models and has had enforcement powers since 2 August 2026.
- National authorities supervise AI systems. In Cyprus, the Commissioner of Communications is the market surveillance authority, the notifying authority and the single point of contact, a centralised model.
- A scientific panel of independent experts can alert the AI Office to systemic risks, and an advisory forum represents stakeholders.
- Each Member State must have at least one AI regulatory sandbox operating by 2 August 2027.
13. Penalties (Article 99)
| Infringement | Maximum fine |
|---|---|
| Using a banned AI practice (Article 5) | €35 million or 7% of worldwide annual turnover, whichever is higher |
| Breaching other obligations, for example of providers, deployers or transparency (Articles 16, 22-26, 50 and others) | €15 million or 3%, whichever is higher |
| Supplying incorrect, incomplete or misleading information to authorities | €7.5 million or 1%, whichever is higher |
For SMEs and start-ups, each cap is the lower of the two amounts. Authorities consider the gravity and duration of the infringement, the harm caused, cooperation, the measures in place and whether it was intentional or negligent.
14. Key dates
- 1 August 2024Entry into force.
- 2 February 2025Banned practices and AI literacy apply.
- 2 August 2025General-purpose AI obligations, governance and penalty rules apply; Member States designate their authorities.
- 2 August 2026Most remaining provisions apply, including transparency (Article 50) and the AI Office's enforcement powers.
- 2 December 2026Ban on non-consensual intimate imagery and child sexual abuse material; deadline for existing generative AI to apply machine-readable marking.
- 2 August 2027General-purpose AI models placed on the market before 2 August 2025 must comply; national sandboxes operational.
- 2 December 2027Requirements for high-risk AI in the Annex III use cases apply.
- 2 August 2028Requirements for high-risk AI in Annex I products, such as medical devices, apply.
- 2 August 2030High-risk AI systems used by public authorities must comply.
The high-risk dates reflect the changes agreed in the 2026 AI Omnibus, as shown on the implementation timeline (last updated 31 August 2026).
15. Official tools and resources
The Future of Life Institute's site artificialintelligenceact.eu offers free tools that help you work through the Act. Each opens in a new tab.
16. What to do now
- List the AI you use or build, including AI features inside software you buy.
- Classify each system: banned, high-risk, transparency only or minimal. Pay particular attention to triage, eligibility, insurance pricing, recruitment and medical-device software.
- Stop any banned use immediately; these rules already apply.
- Train your staff on AI and keep a record (AI literacy).
- For high-risk systems you deploy, get the provider's instructions for use, assign human oversight, keep logs for at least six months and update your data protection impact assessment.
- Add AI disclosures to chatbots and AI-generated content.
- Review your contracts with AI suppliers so they cover documentation, incident reporting and cooperation with authorities.
Sources
- Regulation (EU) 2024/1689 (AI Act) on EUR-Lex
- artificialintelligenceact.eu: high-level summary, implementation timeline, Article 4, 26, 50 and 99 pages, national implementation plans, SME guide, Code of Practice overview, AI therapy guide
Important: this is a summary prepared by Cyprus Connect for general information. It is not legal advice and does not replace the official text, which prevails in case of any difference. Laws and guidance change; check the sources below before relying on a point.