High-risk systems carry the AI Act’s most extensive set of obligations. Originally most of them were to apply from August 2026. With the Digital Omnibus, Regulation (EU) 2026/1744, the EU postponed them: for Annex III systems they apply from 2 December 2027, for AI in regulated products under Annex I from 2 August 2028. A key reason was that the harmonised standards companies are meant to use to demonstrate compliance were not ready in time.
What counts as high-risk
A system is high-risk if it serves as a safety component of a product covered by the product legislation listed in Annex I and requiring third-party conformity assessment, such as machinery or medical devices. Or if it is used in one of the areas in Annex III: biometrics, critical infrastructure, education, employment and workforce management, access to essential private and public services such as creditworthiness assessment or insurance pricing, law enforcement, migration, and justice and democratic processes.
Article 6(3) contains an important exception: an Annex III system is not high-risk if it poses no significant risk, for example because it only performs a narrow procedural task, improves the result of a previously completed human activity or performs a preparatory task. Systems that profile natural persons, however, are always high-risk. Relying on the exception requires documenting the assessment.
Provider obligations
Providers of a high-risk system must run a risk management system across the whole lifecycle, govern training, validation and test data against quality criteria, produce technical documentation, enable automatic logging, provide deployers with understandable instructions for use, enable human oversight by design and achieve an appropriate level of accuracy, robustness and cybersecurity. On top come a quality management system, the conformity assessment, CE marking, registration in the EU database, post-market monitoring and reporting of serious incidents. Providers outside the EU must appoint an authorised representative in the EU.
Deployer obligations
Deployers use the system according to its instructions, assign human oversight to people with the necessary competence and authority, check input data under their control, monitor operation, inform the provider of risks and incidents, keep the automatically generated logs for at least six months and inform workers before a system is used in the workplace. Certain deployers must carry out a fundamental rights impact assessment before first use. People affected by a decision based on a high-risk system have, under certain conditions, a right to an explanation.
Why waiting is expensive
The postponement creates time but no relief on substance. Many requirements concern architecture: logging that can hardly be retrofitted completely, data provenance that cannot be reconstructed for past training runs, human oversight that changes processes. Anyone developing or procuring a system today that will be high-risk from December 2027 should build or select it so that it can meet the requirements then. A rebuild under deadline pressure costs many times more.
A realistic roadmap
1. Inventory and classification: which systems fall under Annex III, which could rely on Art. 6(3)? Our high-risk classification checklist guides through the check.
2. Clarify roles: where is the company provider, where deployer, where does a change of role threaten through own modifications?
3. Gap analysis against the requirements: documentation, data, logging, oversight, robustness.
4. Plan architectural and process measures and fit them into product development.
5. Follow standardisation: once harmonised standards exist, conformity with them creates a presumption of compliance.
6. For deployers: review contracts with providers for documentation, instructions and support, and prepare the FRIA.
Support
For classifying individual systems, AI Risk Check fits. For implementing the requirements we offer guided high-risk conformity and the FRIA and DPIA package. The bigger picture is in our AI Act 2026 guide.

