THE RISK PYRAMID
The AI Act sorts systems into four tiers: unacceptable (banned outright), high-risk (heavily regulated), limited-risk (transparency duties), and minimal-risk (unregulated). Most enforcement weight falls on the high-risk middle — biometric ID, hiring tools, credit scoring, border systems.
THE BRUSSELS EFFECT
The EU has no major frontier AI lab, yet writes the rules the labs follow. GDPR did this for data; the AI Act repeats the pattern. When compliance costs are fixed and the EU market is 450 million consumers, global firms build to the strictest rule and ship it everywhere.
WHY NUDIFIERS ARE A SPECIAL CASE
Nudifier apps generate non-consensual intimate imagery from ordinary photos. Victims are overwhelmingly women and girls; school-age cases have surged across Europe since 2023. The harm pattern fits no existing category cleanly — it is neither classical defamation nor traditional CSAM when the subject is adult — which is why a bespoke ban was politically easier than a court ruling.
THE DELAY TRADE
Industry lobbied hard against the high-risk tier's compliance burden — conformity assessments, fundamental rights impact assessments, post-market monitoring. The 16-month delay on biometrics and border-control rules is the price Parliament paid to get the nudifier ban through. Simplification packages typically swap a high-salience win for a low-visibility concession.
THE ENFORCEMENT GAP
Apps proliferate on infrastructure outside EU jurisdiction — Telegram channels, offshore web hosts, payment rails in third countries. The Act binds providers and deployers, but the enforcement reach against an anonymous developer in a non-cooperative jurisdiction is weak. App stores and payment processors become the de facto chokepoint.