HumanosTech’s November 2026 EU AI Act guide points high-risk teams to December 2027 and August 2028
Key Takeaways
- Treat the Omnibus as sequencing, not suspension. Transparency and basic duties still need owners.
- Start with an AI register that captures systems, suppliers, use cases, and likely risk category.
- Reserve heavier conformity work for Annex III and Annex I systems, but build evidence now.
Why it matters
- ProductProduct leaders need to decide which AI features need transparency work now and which may require high-risk documentation later.
- InvestorsInvestors can assess whether portfolio companies have real AI inventories or only policy documents.
The Digital Omnibus does not pause the AI Act. It changes which files should be built first.
The new EU AI Act calendar has a very unglamorous hero: the spreadsheet. If you were planning for one compliance cliff, HumanosTech’s November 2026 guide makes the more annoying point. The cliff has become a staircase. Annex III high-risk systems now point to December 2027, Annex I product systems to August 2028, and the rest of the Act has not gone on holiday.
That distinction matters because compliance theater usually starts with a policy PDF and ends with nobody knowing which models are actually in use. The Digital Omnibus gives some teams more time for the expensive files. It does not give them permission to stay ignorant.
What moved, and what did not
HumanosTech’s November 2026 guide says the EU AI Act has been in force since 1 August 2024, with prohibitions applying from February 2025, general-purpose AI rules from August 2025, and transparency duties from August 2026. The same guide says the Digital Omnibus pushes high-risk obligations to December 2027 for Annex III systems and August 2028 for Annex I systems. The European Commission’s AI Pact page separately says the AI Office and national authorities took enforcement powers on 2 August 2026, while some high-risk requirements apply only from December 2027.
AI Act dates after Omnibus
- 1 August 2024Act in force
- February 2025Prohibitions
- August 2025GPAI
- August 2026Transparency
- December 2027Annex III
- August 2028Annex I
The practical reading is simple. The Omnibus changes sequencing, not scope. If a vendor tells you the AI Act was delayed, ask which article, which system category, and which obligation. That usually ends the meeting.
Scope is still the first control
The European Commission describes the AI Act as a risk-based framework for AI developers and deployers. Its examples of high-risk uses include critical infrastructure, education access, robot-assisted surgery, and employment tools such as CV-sorting software. That means classification is not a brand exercise. It depends on how the system is used, who relies on it, and whether it affects access to regulated or consequential services.
HumanosTech also says the Act has extraterritorial scope and penalties can reach EUR 35 million or 7% of global turnover. So a non-EU builder serving EU users should not treat the calendar as a Brussels-only housekeeping issue. The first defensible control is an inventory that can separate ordinary internal automation from systems that may fall into Annex III or Annex I. Without that, December 2027 is just a date on a slide.
Article 4 got narrower, Article 50 stayed awake
HumanosTech says the Digital Omnibus amended Article 4 from a duty framed around ensuring a sufficient level of AI literacy to one focused on measures that support it. Khairos reaches the same practical conclusion for SMEs: literacy support remains, transparency duties remain in force, and the heavier technical steps for many high-risk systems move later. Translation: the training deck can be modest, but it should exist, and it should match the tools people actually use.
Article 50 is the part many product teams will feel first, because transparency is already on the calendar according to HumanosTech and Khairos. If users interact with AI or receive AI-generated outputs in contexts covered by the rule, the compliance work is not waiting for Annex III. Builders should review interface notices, output labeling, vendor representations, and logs now. Nobody gets credit for discovering in 2027 that the chatbot has been pretending to be a human since 2025.
The staged program builders should run now
Khairos advises deployers to keep a register with system names, supplier contacts, and use cases. The European Commission’s AI Pact similarly encourages organizations to identify and map AI systems likely to be high-risk. That is not glamorous work, which is a useful sign. Real compliance usually starts where procurement records, product owners, and security teams disagree about what is running.
For Annex III candidates, HumanosTech ties full obligations to EU database registration, conformity assessment, human oversight, and traceability. For Annex I, the guide links the later August 2028 date to AI in regulated products such as machinery, vehicles, and medical devices, with CE marking. Product teams should not write the whole technical file tomorrow if classification is still unresolved. They should assign owners, preserve design evidence, document intended use, and tighten vendor contracts around audit support, change notices, logging, and incident cooperation.
The useful effect of the Digital Omnibus is not comfort. It is ordering. Do transparency and literacy support now, build the register immediately, classify systems before budget season, and reserve the heavier conformity machinery for systems that actually land in high-risk categories. The next thing to watch is how the AI Office and national authorities use their 2 August 2026 enforcement powers, because the first cases will teach builders which paperwork was real and which was just decorative compliance.
Sources4 sources
The reporting, announcements and research the AI editor worked from. Links open the original publisher.
