
What small teams and creators should do now
Share
TL;DR: The AI Act entered into force on Aug 1, 2024. Bans on certain AI practices apply since Feb 2, 2025; GPAI obligations are live since Aug 2, 2025. Most other rules kick in Aug 2, 2026, with some high-risk obligations stretching to Aug 2, 2027. Start with inventory, transparency, and lightweight risk controls.
Why it matters right now
-
The deadlines are fixed—the Commission confirmed the timeline despite calls to delay. Reuters
-
For General-Purpose AI (GPAI), guidance and a voluntary Code of Practice are out to help show compliance. Digitale Strategie Europa+1
-
Fines can reach up to €35M or 7% of global turnover for serious violations. (Stick to proportionate, documented controls.)
Key dates (as of Sept 2025)
-
Feb 2, 2025: Prohibited practices banned. Künstliche Intelligenz Gesetz EU
-
Aug 2, 2025: GPAI obligations begin (with a grace path for models placed on the market before this date—until Aug 2, 2027). European Commission+1
-
Aug 2, 2026: Most other obligations apply. Künstliche Intelligenz Gesetz EU
-
Aug 2, 2027: Additional high-risk/GPAI transition endpoints. Künstliche Intelligenz Gesetz EU+1
Am I affected?
Yes, if you:
-
Use AI features in your app/shop (recommendations, moderation, chatbots).
-
Publish or sell AI-generated content.
-
Train or fine-tune models yourself.
If you’re a GPAI provider (or rely on one), check the new GPAI guidelines to understand what “transparency, safety, and documentation” mean in practice. Digitale Strategie Europa
7-step quick checklist (practical & lightweight)
-
AI inventory
List every AI feature: purpose, inputs, outputs, vendor/model. Keep it to one page per feature. -
Risk notes
Write a short note on misuse, failure modes, bias risks, and how users can report issues. -
User transparency
Tell users where AI is used and how to reach a human when needed. (Simple UI copy is fine.) -
Data hygiene
Minimise personal data, log only what you need for safety/quality, set retention windows. -
Vendor diligence
Ask vendors for model cards/docs; align with the GPAI Code of Practice where possible. Digitale Strategie Europa -
UI failsafes
Add “AI can be wrong” hints, easy report buttons, and opt-outs for personalisation. -
Team literacy
Give your staff a 30-minute brief on AI risks, labeling, and incident reporting.
Copy-paste notice (you can use this verbatim)
AI notice: This feature uses AI and may be inaccurate. Need help or want a human review? Contact support—we’re happy to assist.
Bottom line
You don’t need a legal department to start. A clear feature inventory, a page of risk notes, and transparent UI copy already cover most of the journey—and put you on track for the 2026/2027 milestones. Künstliche Intelligenz Gesetz EU