The EU Commission is going on the offensive: From September 2026, the AI Act will no longer be treated merely as guidance but will be fully enforced. This means concrete action – formal information requests (RFI) are already going out to multiple AI companies, and the Commission reserves the right to audit models and potentially block them. This marks the first binding enforcement phase of Europe's AI rulebook.
Key facts
- AI Act enters full enforcement from September 2026 – first formal information requests already underway
- Federal Digital Minister Karsten Wildberger (CDU) rejects a development moratorium; Germany opts for international cooperation instead
- National AI Safety Institute AISI Germany is being established and will collaborate with institutes in the UK and France
- Dario Amodei (Anthropic) calls for coordinated pause in the West; warns of hundreds of billions in damages from uncontrolled AI development
Wildberger: No halt, but global oversight
While prominent AI chiefs like Dario Amodei, Elon Musk, and Demis Hassabis push for a coordinated slowdown in AI development, the German government is taking a different approach. Digital Minister Karsten Wildberger called a halt to AI development in Europe "critical" and "not a viable path," according to reports from September 13, 2026.
Instead, Germany wants to bring the issue of global AI oversight to G7 talks with the US and China. The focus is on building digital sovereignty and national AI expertise – specifically through the new AISI Germany safety institute, which will work closely with partners in the UK and France.
The Anthropic warning scenario
Amodei sketches a grim scenario in his essay: a swarm of rogue AI agents could seize control of the internet within six to twelve months – causing economic damage in the hundreds of billions of dollars. His solution is a coordinated, industry-wide pause in the Western world for one to two years. This time would be used to establish safety measures through independent auditors and shared standards among democracies, with the goal of international agreements including China.
Amodei receives support from Elon Musk (SpaceX, xAI), Demis Hassabis (DeepMind), and Sam Altman (OpenAI).
Liability and risk tiers tighten
In parallel, the EU Commission is sharpening the rules themselves. The AI Act defines risk categories for AI systems – from minimal to high-risk – and ties liability rules to them. From September, these will no longer be treated as recommendations but as binding requirements.
| Aspect | Previously | From September 2026 |
|---|---|---|
| AI Act Status | Guideline, transition period | Full enforcement |
| Oversight | National authorities | EU Commission + national authorities |
| Sanctions | Planned | Active |
What this means for German companies
This is a wake-up call: the AI Act is no longer a future concern but will be actively enforced from September 2026. German companies developing or deploying AI systems should now review their compliance structures – especially for high-risk applications (recruitment, credit decisions, security). The government is simultaneously building its own safety infrastructure, which also creates opportunities for German AI safety providers. What remains unclear is whether the international cooperation strategy will deliver results quickly enough, or whether calls for a moratorium will gain momentum again.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
