Auditor Lars Zimmermann, ISO/IEC 42001 Senior Lead Auditor (PECB)
Verified at the source
Status of the EU AI Act and the standards
This page records what applies today. Each entry names its source and the day I last checked it against the primary source. When something changes, the entry says what changed and when. Nothing is overwritten silently.
The EU AI Act, Regulation (EU) 2024/1689, applies as amended by Regulation (EU) 2026/1744 of 8 July 2026, the so-called Digital Omnibus. The amendment was published in the Official Journal on 24 July 2026 and has been in force since 27 July 2026. So far this is the only amending regulation.
Change: Amended on 27 July 2026 by Regulation (EU) 2026/1744.
Providers and deployers take measures to support the development of AI literacy of their staff and of other persons acting on their behalf. They do not have to guarantee a particular level for anyone. The article names no number of hours, no certificate and no specific training. Article 4 is not on the list of fines in Article 99(4).
Change: Reworded on 27 July 2026. The earlier wording required best efforts to ensure a sufficient level of AI literacy.
The prohibitions have applied since 2 February 2025. Two further prohibitions apply from 2 December 2026: AI that generates or manipulates realistic intimate depictions of an identifiable person without their consent, and AI for child sexual abuse material. Infringements can be fined up to EUR 35 million or 7 per cent of annual turnover.
Change: The two new prohibitions were inserted by Regulation (EU) 2026/1744.
AI that infers the emotions of people in the workplace or in education institutions is prohibited, except for medical or safety reasons. By definition, an emotion recognition system identifies or infers emotions or intentions on the basis of biometric data, such as voice or facial expressions. Analysing written text alone does not fall under this definition. That does not make it lawful automatically: data protection and co-determination rules still apply.
The obligations for high-risk AI systems apply from 2 December 2027 for Annex III systems, for example in employment, education or credit, and from 2 August 2028 for AI in regulated products under Annex I.
Change: Postponed on 27 July 2026. Before that the dates were 2 August 2026 (Annex III) and 2 August 2027 (Annex I).
AI used exclusively for non-safety-related aspects of user support, performance optimisation, automation, usability or quality control is not a safety component. If its failure or malfunction would endanger health and safety, it is a safety component all the same. Since the amendment the Machinery Regulation is listed in Annex I Section B.
Change: Inserted on 27 July 2026 by Regulation (EU) 2026/1744.
The transparency obligations have applied since 2 August 2026; the date was not postponed. The notice for chatbots and the machine-readable marking of artificially generated content are owed by the provider. Deployers must disclose deepfakes and inform people about emotion recognition or biometric categorisation. For systems placed on the market before 2 August 2026, the period for marking under paragraph 2 runs until 2 December 2026. Infringements are on the list of fines in Article 99(4).
Deployers assign human oversight to natural persons who have the necessary competence, training and authority. Where they have reason to consider that use in line with the instructions may result in a risk to health, safety or fundamental rights, they inform the provider or distributor and the market surveillance authority and suspend the use. The regulation does not require an AI officer anywhere.
Small mid-cap enterprises are included in several reliefs. For fines under Article 99(4) and (5) the lower amount applies to them, but not for infringements of the prohibitions in Article 5. Priority access to national AI regulatory sandboxes remains reserved for SMEs including start-ups. The AI Office may additionally set up a sandbox at Union level.
Change: Article 99(6a) and Article 57(3a) were inserted by Regulation (EU) 2026/1744.
The rule on when special categories of personal data may be processed to detect and correct bias is in Article 4a. Article 10(5), where it used to be, has been deleted.
Change: Moved on 27 July 2026 by Regulation (EU) 2026/1744.
So far a single harmonised standard is listed: EN 18286:2026 on the quality management system of providers, published with Implementing Decision (EU) 2026/2274 in the Official Journal of 14 October 2026. ISO/IEC 42001 is not listed as a harmonised standard. A certificate under ISO/IEC 42001 therefore creates no presumption of conformity under the EU AI Act.
Change: EN 18286:2026 has been listed since 14 October 2026.
The competent market surveillance authority is the Federal Network Agency (Bundesnetzagentur) unless the act provides otherwise. It hosts an independent AI market surveillance chamber, responsible for high-risk systems in law enforcement, border management, migration, justice and democratic processes, and for biometric systems where they are used in those areas. Sector-specific responsibilities remain in place.
The current edition is ISO/IEC 42001:2023. The standard is certifiable. Annex A contains 38 controls in nine topic areas (A.2 to A.10), binding via the Statement of Applicability. Annex B is also labelled normative and gives implementation guidance. An organisation may adapt that guidance or replace it with its own implementation and does not have to justify excluding it (B.1). Human oversight appears in that guidance; Table A.1 has no separate control for it.
ISO/IEC 42006:2025 sets minimum times for the initial audit, by the number of people in the AI life cycle and by role. In the smallest band of up to ten people it is 3.5 auditor days for organisations that provide or use AI, 5 days for organisations that develop AI and 6.5 days for multiple roles. The values rise with headcount and are minimums before adjustments.
ISO 9001:2026 is the sixth edition, published in September 2026. It replaces ISO 9001:2015. Clause 10 now has only two subclauses: 10.1 Continual improvement and 10.2 Nonconformity and corrective action. In 9.2.2 the selection of auditors is now under item b. The word certificate and a three-year period for internal audits do not appear in the standard.
Change: New edition since September 2026. Anyone who refers to the 2015 edition by number should check their cross-references.
ISO 19011:2026 is the fourth edition, published in May 2026. It replaces ISO 19011:2018. The standard gives guidelines for auditing management systems and contains no requirements. The main changes are expanded guidance on remote audits and on virtual locations.
ISO/IEC 27090:2026 is the first edition, published in October 2026. It is guidance and not certifiable. Clause 6 describes nine AI-specific types of attack, from data poisoning to prompt injection. Annex B lists six conventional attacks in an AI-specific form, including supply chain attacks.
The International Accreditation Forum (IAF) ceased operations on 1 January 2026. Mutual recognition is continued by the Global Accreditation Cooperation (Global ACI), and existing accreditations remain recognised. Global ACI continues the former IAF mandatory documents under its own numbers, to be applied since 28 August 2026, for example Global ACI-TECH-3-001 for multi-site organisations.
Change: Global ACI instead of IAF since 1 January 2026.
Directive (EU) 2024/2853 counts software as a product. It applies to products placed on the market or put into service after 8 December 2026. The German act modernising product liability law has been promulgated (Federal Law Gazette 2026 I No. 321 of 27 October 2026) and enters into force on 9 December 2026. There is no separate AI liability directive; the Commission withdrew the proposal in 2025.
Change: Cut-off date set to 8 December 2026 by corrigendum of 7 May 2026. Act promulgated on 27 October 2026.
The implementing act has been in force since 6 December 2025. In the sectors covered it applies to entities with at least 50 employees or with annual turnover and an annual balance-sheet total of more than EUR 10 million each. A significant security incident requires an early warning within 24 hours, a notification within 72 hours and a final report one month after that.
Checks are made against the wording: the Official Journal of the EU, the consolidated version, the official German statute portal and the standards themselves. Third-party summaries do not count as evidence.
The verification date applies per entry. An older date means that point has not been checked against the source again since then.
Standards are cited by clause number and in my own words, not verbatim.
This is a professional assessment by an auditor and not legal advice.
How to cite this page
Der KI-Auditor (Lars Zimmermann): Status of the EU AI Act and the standards, entry with anchor, verification date of the entry, der-ki-auditor.de/en/eu-ai-act-status/.
You may use, adapt and share the texts on this page freely, including commercially. The condition is attribution to “Der KI-Auditor, der-ki-auditor.de”. Licence: CC BY 4.0.