AI Ethics

The EU AI Act One Year Later: What's Working, What's Failing

By Priya Patel · May 7, 2026 · 11 min read
The EU AI Act One Year Later: What's Working, What's Failing

The world's first comprehensive AI law has been in effect for 12 months. We analyze the impact on startups, big tech, and consumer protection.

What's Working

  • Transparency requirements forcing AI companies to disclose training data sources
  • Consumer protection has measurably improved — fewer dark patterns in AI products
  • High-risk AI systems (hiring, credit scoring) now require human oversight by law
  • Watermarking of AI-generated content is becoming standard across the EU

What's Failing

  • EU AI startups report a 30% innovation slowdown vs US peers
  • Compliance costs are crushing small AI companies (€100k+ per certification)
  • Definitions of "high-risk" are vague — companies err on side of over-compliance
  • Enforcement is wildly inconsistent across member states

The Verdict

The intent was right — protect citizens from AI harm. The execution has been heavy-handed enough to push talent and capital to the US and Asia. Expect amendments in 2027.