#97 AI security and governance 5.4 min
EU AI Act for Developers: A Practical Tier Checklist
Before you read the EU AI Act, one question decides most of it: which tier is your app in?
Make the spam filter produce content
Copy the code folder first. In apps.py change makes_content to True for the spam filter and run compare.py. Predict which tier the spam filter moves to and which duty it gets. You should see it become limited with the duty to mark generated content as AI-made, and the totals become 2 high risk, 3 limited and 1 prohibited, with no minimal app left. Change makes_content back to False afterwards.
A hint
Look at the last two tests in the tier function in tiers.py.
Remove the high risk area
In apps.py change high_risk_area from 'employment' to None for the CV screener and run compare.py. Predict its tier and the totals. You should see the CV screener become minimal, and the totals become 2 minimal, 2 limited, 1 prohibited and 1 high risk (this assumes the spam filter is back to its original setting). Change high_risk_area back to 'employment' afterwards.
A hint
The high risk test only looks at whether high_risk_area is filled in.
Add a seventh app that only talks
Add a new app to APPS in apps.py that does not make content, is not banned and has no high risk area, but does talk to people. Give it any name. Run compare.py. You should see the app limited, with the duty to tell people they are talking to an AI as its only duty. Remove the new app again afterwards.
A hint
Use banned=False, high_risk_area=None, talks_to_people=True and makes_content=False.
Ban a high risk app
With apps.py back to the original, set banned to True for the CV screener, which is also in the employment area. Predict its tier before you run compare.py. You should see the CV screener become prohibited, and the totals become 2 prohibited, 2 limited, 1 minimal and 1 high risk. Then explain why, using the order of the tests in tier().
A hint
The test that comes first wins.
These tasks are for you to try on your own machine. Nothing is checked or saved.
What this does not show
Not legal advice. Six made-up apps and four yes or no questions; the rules do not decode the Annex III list, the exceptions, or the difference between a provider and a deployer. Dates are as shown on the European Commission page in November 2026 and have already changed once; check the current text.
Up next in AI security and governance
More in AI security and governance
- #18 Prompt Injection Live: Attacking My Own LLM App
- #19 Redact Personal Data Before Your LLM Sees It (Python)
- #88 Indirect Prompt Injection: A Note That Gave Orders
- #89 Jailbreak-Test Your Own App: Eight Wrappers, Counted
- #90 Stop Leaking API Keys: Build a Secret Scanner in Python
- #91 Rate Limiting an LLM API: Token Bucket in FastAPI
Related videos
- #214 Build an AI Model Register and Risk Score Data governance
- #229 A Reference Architecture for Enterprise AI Agents, as Code Enterprise architecture and AI agents
- #248 TOGAF and ArchiMate: Where an AI Agent Fits, as Code Enterprise architecture and AI agents
- #100 Fake but Realistic Test Data in Python With Faker AI security and governance
