DEV Community

Cover image for Is your AI wrapper a "High-Risk" system? (A dev's guide to the EU AI Act)
Damir
Damir

Posted on

Is your AI wrapper a "High-Risk" system? (A dev's guide to the EU AI Act)

If you're building AI features right now, you and your team are probably arguing about the tech stack:

  • Should we use LangChain or LlamaIndex?
  • Should we hit the OpenAI API or run Llama 3 locally?

Here is the harsh truth about the upcoming EU AI Act:

Regulators do not care about your tech stack.

They don’t care if it’s a 100B parameter model or a simple Python script using scikit-learn.

The law only cares about one thing:

Your use case.


Why This Matters

Your use case determines your risk category.

If your product falls into the High-Risk category, you are legally required to implement:

  • human oversight
  • risk management systems
  • detailed technical documentation (Annex IV)

Getting this wrong doesn’t just mean “non-compliance”.

It means:

  • failed procurement audits
  • blocked enterprise deals
  • serious regulatory exposure

🔍 5 Real-World AI Scenarios

Here are practical examples to help you understand where your system might fall.


1. AI Chatbot for Customer Support

Use case:

  • routing tickets
  • answering FAQs

Classification:

👉 Limited Risk

Dev requirement:

Add UI elements disclosing that users are interacting with AI.

The trap:

If your bot starts making decisions (e.g. auto-refunds, banning users), you might cross into High-Risk territory.


2. AI for CV Screening / Hiring

Use case:

  • parsing resumes
  • ranking candidates

Classification:

👉 High-Risk (explicitly listed under Annex III)

Dev requirement:

  • bias monitoring
  • human-in-the-loop (HITL) flows
  • full decision logging

3. E-commerce Recommendation Engine

Use case:

  • tracking user behavior
  • suggesting products

Classification:

👉 Minimal Risk

Dev requirement:

Almost none under the AI Act (GDPR still applies).


4. AI Credit Scoring System

Use case:

  • determining loan eligibility

Classification:

👉 High-Risk

Dev requirement:

Full traceability — you must be able to explain decisions made by the system.


5. AI Generating Marketing Content

Use case:

  • generating blog posts
  • writing ad copy

Classification:

👉 Minimal to Limited Risk

Dev requirement:

Minimal — unless generating deepfakes (then disclosure/watermarking applies).


🛠️ The Real Risk: Feature Creep

The biggest danger isn’t writing documentation.

It’s this:

Your system can move from Limited Risk to High-Risk with a single merged PR.

A small feature change can completely change your regulatory obligations.


Quick Self-Check

If you’re targeting the EU market, ask yourself:

  • Does my system influence hiring decisions?
  • Does it impact financial outcomes?
  • Does it affect people’s rights or opportunities?

If yes:

👉 You may already be in High-Risk territory.


🧪 A Simple Way to Check

If you're not sure, I built a free developer tool to calculate this instantly:

👉 https://www.complianceradar.dev/ai-act-risk-classification

No signup required.


Final Thought

Most AI products won’t fail because of bad code.

They’ll fail because of misunderstood regulation.

Understand your risk level early — and build with confidence.


💬 What kind of AI features are you building right now?

Drop your use case below and we can try to classify it together.

Top comments (0)