What is changing — and what is taking shape before it applies. We check the official sources every 14 days and write down what affects companies that use AI. Every entry dated and linked to the primary source, so you can check it yourself.
Action needed
Germany: AI supervision now has a name
On 29 July 2026, Germany's AI Market Surveillance and Innovation Promotion Act (KI-MIG) entered into force. It forms Article 1 of the Act implementing Regulation (EU) 2024/1689, published in the Federal Law Gazette 2026 I No. 223.
It is now settled who supervises the AI Regulation in Germany. The Bundesnetzagentur (Federal Network Agency) is the competent market surveillance authority unless the Act provides otherwise (section 2(1) KI-MIG); sector-specific authorities such as the financial supervisor remain in charge of their fields. Until now this was only announced; it is now law. The Bundesnetzagentur also operates a complaints office. In practice, proceedings usually begin with a complaint rather than an inspection on the authority's own initiative.
The Act does not change your substantive obligations. The transparency obligations under Article 50 continue to apply as they have since 2 August 2026. The obligations for high-risk systems remain postponed, to 2 December 2027 and 2 August 2028 respectively. What is new is a competent authority, not a new requirement for you.
One point is worth checking. The same Act extended the German Whistleblower Protection Act: breaches of the AI Regulation are now expressly reportable. If you have 50 or more employees and therefore operate an internal reporting channel, that channel must accept and process such reports. You do not need a new system for this. It is enough to add the case categories to your reporting channel and inform the people responsible for it. If you have fewer than 50 employees, nothing changes for you.
Better tools are now available for implementing Article 50. The European Commission published its final guidelines on 20 July 2026, supplemented by a voluntary Code of Practice on marking AI-generated content dated 10 June 2026. Neither is legally binding, but both are useful for orientation. The part on machine-readable marking is addressed to the providers of AI tools, such as OpenAI or Google, not to you as a user. For you, a clear and comprehensible label where one is required remains sufficient. There is still no official EU seal for AI.
This message concerns Germany. Nothing changes as a result for Austria or the Netherlands. We will inform you about the Dutch implementing act once it has been adopted.
Good to know
What comes next: the dates until the end of 2026
Two dates are still ahead this year. On 2 December 2026 new prohibitions take effect, and the grace period for machine-readable marking of legacy systems ends — that duty, however, falls on the providers of the AI tools, not on you as a user. On 9 December 2026 the transposition deadline for the new EU product liability directive expires; software and AI systems expressly count as products there. Nothing needs doing on your side at present. We will get in touch as soon as one of these dates becomes concrete for your organisation.
Action needed
Since 2 August 2026: labelling and market surveillance
The transparency duties under Article 50 have applied since 2 August 2026, and national market surveillance has begun its work. Unlike Article 4, Article 50 carries a fine directly: up to 15 Mio euros or 3 percent of turnover, whichever is lower for small and medium-sized companies. What you need is a written rule on how your organisation labels AI content, and staff who know it. Ordinary marketing copy, quotes and product descriptions, by the way, do not require labelling.
Legal position changed
Adopted: high-risk duties postponed, transparency stays
The Council gave final approval to the Digital Omnibus on 29 June 2026, and it has been in force since July 2026. The obligations for high-risk systems were postponed: standalone systems to 2 December 2027, systems inside regulated products to 2 August 2028. The transparency duties and the start of national market surveillance were not postponed. For companies that merely use AI, that is the more relevant half, since hardly any of them build high-risk systems themselves.
Good to know
Proposal: Commission plans to postpone deadlines (Digital Omnibus)
On 19 November 2025 the European Commission proposed postponing parts of the AI Act. The proposal is called the Digital Omnibus. Nothing is settled yet: Parliament and Council have to agree, and proposals change during the procedure. We are telling you today because a possible postponement affects your planning. Nothing needs doing at this point.
Good to know
Article 4 applies: AI literacy is mandatory
Article 4 of the EU AI Act has applied since 2 February 2025. Companies whose staff use AI must foster their AI literacy. The prohibitions in Article 5 took effect the same day, among them emotion recognition in the workplace. Two things follow for you: trained staff, and a written rule stating what is allowed and what is not.
Wissen ist die eine Hälfte. Der Nachweis ist die andere.
Diese Seite sagt Ihnen, was sich geändert hat. Im Abo erfahren Sie zusätzlich, was Sie tun müssen — per E-Mail, ohne dass Sie nachsehen müssen, mit Kenntnisnahme-Protokoll für Ihre Nachweis-Akte. Genau das fragt ein Prüfer oder Geschäftskunde ab.
Tarife ansehen →
Diese Seite ist eine Einordnung nach bestem Wissen und keine Rechtsberatung. Verbindlich ist immer der Wortlaut der jeweiligen Rechtsquelle — deshalb steht bei jeder Meldung der Link dorthin. Fragen an kontakt@klarcomply.com.