On June 11, 2026, the Bundestag passed the AI Implementation Act (BT-Drs. 21/6407). From August 2, 2026, the Federal Network Agency (Bundesnetzagentur, BNetzA) becomes the central market surveillance authority for AI systems in Germany. For SMBs — whether a trades business with an AI phone assistant or a financial advisor with automated lead qualification — that means: transparency obligations, documentation, and risk assessment are mandatory, not optional. The good news: an initial assessment covers 80% of it. Here are the three questions you need for that.
What did the Bundestag pass on June 11?
The AI Implementation Act was passed in the version amended by the Digital Committee — CDU/CSU and SPD in favor, AfD, the Greens, and the Left against. Approval by the Bundesrat is still pending.
The core: the Federal Network Agency (BNetzA) becomes the supervisory authority for the EU AI Act. It gets:
- an independent AI market surveillance chamber for high-risk systems
- a coordination and competence center for AI
- a central point of contact and complaints office for citizens
- at least one AI regulatory sandbox
In addition to the EU fines, the BNetzA can impose fines of up to €50,000 for violations of cooperation obligations. The federal government's annual implementation cost is estimated at around €15.9 million.
What applies when? The EU AI Act's phased timeline
The EU AI Act (Regulation (EU) 2024/1689) applies in phases under Art. 113:
| Date | What applies |
|---|---|
| 02/02/2025 | Prohibited practices (Art. 5) + AI literacy obligation (Art. 4) |
| 08/02/2025 | GPAI obligations + governance |
| 08/02/2026 | Transparency obligation (Art. 50) + national market surveillance + enforcement |
| 08/02/2027 | Full application of high-risk rules (Annex I/III) |
August 2, 2026 is the decisive date for most SMBs. From then on, the transparency obligation under Art. 50 becomes enforceable by the BNetzA. Anyone using an AI system that interacts with people must disclose that it's an AI.
What penalties actually apply?
The full fine tiers under Art. 99 of the EU AI Act:
| Tier | Violation | Maximum fine |
|---|---|---|
| 1 | Prohibited practices (Art. 5) | up to €35 million or 7% of global revenue |
| 2 | High-risk obligations | up to €15 million or 3% of global revenue |
| 3 | False information to authorities | up to €7.5 million or 1% of global revenue |
Important for SMBs: Art. 99(6) contains an SMB clause. For SMBs and start-ups, the lower of the two values applies, and proportionality must be taken into account. A trades business with €2 million in revenue and an AI phone assistant (limited risk) risks at most 1% of its revenue for an Art. 50 violation — not €35 million. Manageable, if you're prepared.
Three questions — 30 minutes — a clear assessment
1. Does your AI interact directly with people?
If yes: the Art. 50 transparency obligation applies from 08/02/2026. Your system must disclose at every interaction that it's an AI — a clear greeting announcement on the phone or a note in the chat is enough.
2. Does your AI make decisions with a significant effect on people?
Assessing creditworthiness, pre-selecting job applicants, scoring insurance risk — that's high risk under Annex III. Booking appointments, answering calls, qualifying leads — generally limited risk. The line runs along the specific use case, not the industry.
3. Do you have documentation you could show the BNetzA?
Not a 200-page document. Which system do you use? What data does it process? What decisions does it make? Who is responsible? If you can answer that on two pages, you're prepared for 80% of the requirements. If you suspect an Annex III case, it's worth consulting a specialized law firm.
Core answer: Most SMBs fall under Art. 50 (limited risk). A greeting announcement and a two-page document are enough. No 200-page audit. No legal project.
What does this mean for trades businesses?
An AI phone assistant falls under limited risk under Art. 50: the AI interacts with callers, so it must identify itself as AI. Not a high-risk system, as long as the assistant answers calls, books appointments, and passes on information — but doesn't make creditworthiness decisions or personnel selections.
According to Bitkom's 2026 study "Digitalization of the Trades," only 4% of trades businesses actively use AI. 43% offer no AI training, and 32% believe they're not affected by the AI Act at all. Whoever gets started now does so with clean compliance documentation from day one — instead of retrofitting it later.
More on AI agents for trades businesses — what a compliant setup looks like from day one.
What does this mean for financial advisors?
Here it gets more nuanced. Lead qualification — triaging between existing and new customers, scheduling, initial information — is generally limited risk (Art. 50). The transparency announcement is enough.
As soon as the AI makes a risk-based advisory or acceptance decision with a significant effect — for example, assessing creditworthiness or scoring insurance risk — high risk under Annex III No. 5(b) can apply. Then GDPR Art. 22 (prohibition of purely automated individual decisions) and Art. 35 (data protection impact assessment) come on top.
The rule of thumb: assess per use case, not by industry as a whole. A financial advisor using AI for scheduling and availability faces the same compliance bar as a tradesperson.
Details on AI agents for financial advisors — compliant from day one.
How Ben meets Art. 50 out of the box
Our AI call-answering agent Ben greets every caller with an announcement that makes clear: this is an AI-powered system speaking. That's not a bolted-on compliance measure — it's been part of the setup from day one.
For trades businesses and financial advisors working with Amazing Agents, the Art. 50 transparency obligation is therefore already met as of August 2, 2026. No extra effort, no retrofitting.
We build setups where SMBs can carry compliance responsibility in a controlled, audit-ready way. Clean documentation, a transparent announcement, clear responsibilities — making compliance manageable, not something to fear.
Frequently asked questions about the EU AI Act
Does the EU AI Act apply to my trades business?
If you use an AI system that interacts with customers or suppliers — yes. From August 2, 2026, the transparency obligation (Art. 50) applies: everyone must know they're speaking with an AI. For AI phone assistants like Ben, a greeting announcement is enough.
Is my AI phone assistant a high-risk system?
Generally not. Answering calls, booking appointments, and passing on information falls under limited risk. High risk only arises if the AI makes decisions with a significant effect — for example, assessing creditworthiness or pre-selecting job applicants.
What does compliance cost SMBs?
Bitkom estimates the annual compliance costs for the German economy at up to €20 billion. For a single SMB with an AI phone assistant, the effort is manageable: set up a transparency announcement, document the use case, define responsibilities.
Do I need to bring in a lawyer?
Only if your AI makes decisions with a significant effect (Annex III). For most SMBs, the 3-question self-check and a two-page document are enough.
Does the EU AI Act replace the GDPR?
No. The GDPR remains in force in parallel — with its own obligations (Art. 5, 6, 22, 28, 35) and its own fines (up to €20 million / 4% of revenue). Both frameworks must be complied with at the same time.
Sources:
1. Bundestag text archive KW 24/2026 — AI Implementation Act, 2026-06-11 — bundestag.de
2. Draft legislation BT-Drs. 21/4594, Federal Government, 2026-03-09 — dserver.bundestag.de
3. EU AI Act Art. 99 — Sanctions — artificialintelligenceact.eu
4. Bitkom — German implementation of the AI Act is coming, 2026-06-11 — bitkom.org
5. openPR / Bitkom study data — AI competence in companies, 2026 — openpr.de
6. BMDS — AI Implementation Act legislative process — bmds.bund.de
7. EUR-Lex — Regulation (EU) 2024/1689 — EU AI Act, 2024-07-12 — eur-lex.europa.eu
8. secjur — EU AI Act fines & sanctions calculation, 2026 — secjur.com