AI Act Compliance Obligations from August 2026: What General-Purpose and High-Risk Providers Must Deliver

Regulation (EU) 2024/1689, the AI Act, entered into force on 1 August 2024. The bulk of its obligations become applicable on 2 August 2026, with the prohibitions in Article 5 already binding since 2 February 2025 and the general-purpose AI provisions since 2 August 2025. From 2 August 2026, providers of high-risk systems, deployers in regulated sectors, importers and distributors face a connected set of duties: risk management, technical documentation, post-market monitoring, serious incident reporting, and registration in the EU database under Article 71.

This article translates the second wave of obligations into an operational checklist. It explains who is in scope, which documents must exist on the day of an inspection, which deadlines run from the moment of knowledge, and how a compliance officer can convert the regulation into recurring tasks rather than a one-off project.

The AI Act distinguishes providers, deployers, importers, distributors and authorised representatives. From 2 August 2026, Chapter III on high-risk systems, Chapter IV on transparency, Chapter VII on governance and Chapter IX on post-market monitoring become applicable to all of them. Article 6 defines high-risk systems through two routes: Annex I lists Union harmonisation legislation (medical devices, machinery, toys, civil aviation) where AI is a safety component; Annex III lists stand-alone use cases including biometric identification, critical infrastructure, education access, employment selection, essential private and public services, law enforcement, migration and administration of justice.

Deployers carry a smaller but real package: human oversight under Article 14, input data governance under Article 26, fundamental rights impact assessments for public bodies and certain private deployers under Article 27, and information duties towards affected persons under Article 26(11). Importers and distributors must verify the CE marking, declaration of conformity and technical documentation before placing a system on the Union market. The roles overlap when an organisation modifies a high-risk system substantially, which under Article 25 turns a deployer into a provider. A clear role inventory, maintained by a designated Compliance officer, is the first artefact that an authority will ask for.

Article 11 requires a technical documentation file for every high-risk system, drawn up before market placement and kept up to date. Annex IV specifies its minimum content: a general description of the system, its intended purpose, the persons developing it, version information, hardware and software interactions, instructions for use, a detailed description of the elements of the system and its development process, monitoring, functioning and control, risk management documentation, lifecycle management, data and data governance practices, the chosen accuracy and cybersecurity measures, and a copy of the EU declaration of conformity.

The file is not a marketing brochure. It must allow a notified body or a market surveillance authority to assess conformity with the requirements of Section 2 of Chapter III, which include risk management (Article 9), data governance (Article 10), transparency and information to deployers (Article 13), human oversight (Article 14), and accuracy, robustness and cybersecurity (Article 15). For SMEs, Article 11(1) third subparagraph allows a simplified format, which the Commission is empowered to specify by implementing act. Practically, the file lives in a controlled workspace with versioned attachments, signed approvals and a clear owner. Audit-fest, dokumentiert, § 11-fest.

Article 9 establishes a risk management system that is iterative, runs throughout the entire lifecycle and requires regular review. Identified risks must be analysed, evaluated and addressed by design measures, mitigation and information to deployers. Residual risks must be communicated. The system must consider risks that may emerge when the high-risk AI system is used in accordance with its intended purpose or under conditions of reasonably foreseeable misuse, and risks to vulnerable groups including children.

Article 10 sets data and data governance requirements for systems trained with data. Training, validation and test datasets must be relevant, sufficiently representative, and to the best extent possible free of errors and complete. Datasets must consider the geographical, contextual, behavioural or functional setting of the intended use. Bias detection and correction measures are required, including the lawful processing of special categories of personal data under Article 10(5) where strictly necessary to detect and correct bias. These obligations do not stop at deployment: post-market monitoring under Article 72 feeds back into the risk register, and corrective action under Article 20 must be triggered if a high-risk system no longer conforms. A pre-built risk register and dataset evidence template, mapped to Articles 9 and 10, removes most of the friction.

Article 14 requires human oversight measures appropriate to the risks, level of autonomy and context of use. Oversight must enable the natural persons in charge to fully understand the capacities and limitations of the high-risk system, to monitor its operation, to remain aware of the possible tendency to automatically rely or over-rely on the output (automation bias), to correctly interpret outputs, to decide not to use the system or to override or reverse an output, and to intervene or interrupt operation through a stop button or similar procedure.

For deployers, Article 26 codifies operational duties: use systems in accordance with the instructions for use, assign human oversight to qualified persons, ensure that input data is relevant and sufficiently representative, monitor operation and report serious incidents to the provider and the market surveillance authority, keep automatically generated logs for at least six months unless other Union or national law provides otherwise, and inform workers and their representatives before putting a high-risk system into service in the workplace under Article 26(7). Public bodies and certain private deployers must also conduct a fundamental rights impact assessment under Article 27 and notify the market surveillance authority of its results. Transparency duties under Article 50 add labelling and disclosure requirements for emotion recognition, biometric categorisation, deep fakes and AI-generated text on matters of public interest.

Article 73 introduces a serious incident reporting regime that runs in parallel to existing reporting duties under medical device, GDPR and NIS-2 law. A serious incident is defined in Article 3(49) as any incident or malfunctioning of an AI system that directly or indirectly leads to the death of a person or serious harm to health, a serious and irreversible disruption of the management or operation of critical infrastructure, an infringement of Union law intended to protect fundamental rights, or serious harm to property or the environment.

Providers must report to the market surveillance authority of the Member State where the incident occurred immediately after the provider has established a causal link between the system and the incident, or the reasonable likelihood of such a link, and in any event not later than 15 days after becoming aware of the incident. The deadline is two days for widespread infringements and serious incidents under Article 3(49)(b) involving critical infrastructure, and ten days in case of death. Frist läuft ab Kenntnis. The provider must also cooperate with the authority and conduct the necessary investigation. A practical reporting workflow, similar to the NIS-2 24-hour and 72-hour pathway, prevents missed deadlines: a single inbox, a triage rule, and an evidence trail with timestamps. The information security officer typically owns the alarm chain, with the compliance officer owning the legal classification.

Article 43 governs conformity assessment for high-risk systems. For most Annex III use cases, providers may follow an internal control procedure based on Annex VI, provided the harmonised standards or common specifications adopted under Article 40 or 41 are applied in full. For biometric systems under Annex III point 1, where harmonised standards or common specifications do not exist or are applied only in part, a notified body must be involved under Annex VII. For Annex I products, conformity assessment follows the relevant sectoral legislation, complemented by the AI Act requirements.

Once conformity is established, the provider draws up an EU declaration of conformity under Article 47, affixes the CE marking under Article 48, and registers the system in the EU database under Article 71. The database is publicly accessible for certain fields and run by the Commission. Annex VIII specifies the registration content: provider identity, system name and description, status, instructions for use, intended purpose, components, software versions and member states of operation. Deployers that are public authorities or Union institutions also register their use of high-risk systems under Article 49(1a). Registration before placing the system on the market or putting it into service is a precondition, not a follow-up task. A missing database entry is a documented infringement under Article 99(4)(g).

Chapter V on general-purpose AI models has been applicable since 2 August 2025 for new models, with a transitional regime until 2 August 2027 for models placed on the market before that date. Article 53 requires providers of general-purpose models to draw up and keep up-to-date technical documentation, including training and testing information, to make information available to downstream providers that integrate the model, to put in place a policy to comply with Union copyright law, and to publish a sufficiently detailed summary of the content used for training, according to a template provided by the AI Office.

Article 55 adds duties for general-purpose models with systemic risk, defined in Article 51 by computational thresholds (currently 10^25 FLOPs) or by Commission decision. These models require model evaluation including adversarial testing, systemic risk assessment and mitigation, serious incident reporting to the AI Office, and adequate cybersecurity protection. The Code of Practice prepared under Article 56, the AI Office guidelines under Article 96, and the harmonised standards in development by CEN-CENELEC JTC 21 are the operational reference points. Even organisations that only deploy a general-purpose model carry residual duties under Article 25(2), particularly when fine-tuning shifts them into a provider role for the resulting high-risk system.

Article 99 sets the penalty framework. Non-compliance with the prohibited practices in Article 5 is sanctioned with administrative fines of up to 35 million euro or, if the offender is an undertaking, up to 7 percent of its total worldwide annual turnover for the preceding financial year, whichever is higher. Non-compliance with obligations of providers, deployers, notified bodies, importers, distributors and authorised representatives, other than those in Article 5, is sanctioned with up to 15 million euro or 3 percent of worldwide turnover. The supply of incorrect, incomplete or misleading information to notified bodies and authorities is sanctioned with up to 7,5 million euro or 1 percent of worldwide turnover.

For SMEs including start-ups, the lower of the two amounts applies. Article 100 introduces fines for Union institutions, bodies and agencies. National competent authorities are designated under Article 70 and report to the AI Office and the Board. Market surveillance powers follow Regulation (EU) 2019/1020, with sector-specific extensions in Annex X (data access for law enforcement and migration uses). Andere führen Compliance wie einen Aktenschrank. Wir führen sie wie Software. The cost of a fine is one variable; the cost of a recall, a market-surveillance order or a court injunction is the larger one. For Germany, the supervisory architecture is being implemented through the AI Act Implementation Act (KI-Umsetzungsgesetz), with BNetzA and the Datenschutzkonferenz playing central roles.

CIVAC is a Compliance-Plattform und Officer-as-a-Service. The dual model is straightforward: license the workspace for your internal officers, or have our officers appointed under a Bestellurkunde. Either way, the AI Act becomes a recurring workflow rather than a project. The workspace ships with audit templates mapped to Articles 9, 10, 11, 14, 17, 26, 27, 47, 71 and 73, an evidence vault with EU data residency, role inventories per system, and a serious incident reporting pathway aligned with the 15-day, ten-day and two-day deadlines.

For organisations with overlapping duties under NIS-2 and ISO/IEC 27001:2022, the same workspace handles the 24/72-hour NIS-2 reporting flow and the ISMS evidence trail, avoiding parallel toolchains. Lizenzieren Sie den Workspace für Ihre internen Beauftragten, oder lassen Sie unsere Beauftragten bestellen. The CIVAC SLA for officer appointment is two working days, against two to six weeks in the classical market. Aus dem Lesen einen Auftrag machen. Reach the team at info@civac.de or through the contact form on civac.de. Bestellurkunde, unterschrieben, abgelegt, belegbar. Turn reading into a mandate.