EU AI Act: Big Models Get a Big Checklist (August 2025)

Estimated reading time: 4 min

If AI had birthdays, August 1, 2025, would be its most "unforgettable" one. One year after the EU AI Act entered into force, the grace period for **General-purpose AI (GPAI) models** has officially ended.

"In the past, you just wrote 'Don't trust me' in a disclaimer. Now, you have to explain to Brussels exactly how you were trained, how much power you burned, and whether you peeked at copyrighted data."

GPAI: Welcome to Regulatory Reality

Starting today (August 2, 2025), heavyweights like GPT-4, Claude, and Gemini are entering the era of strict oversight in Europe.

Transparency Obligations: Detailed technical documentation and summaries of training data are now mandatory. Copyright compliance is no longer a suggestion.
Systemic Risk Monitoring: Models with systemic risks must undergo rigorous evaluations and adversarial testing.
Energy Disclosure: Yes, even your carbon footprint during training needs to be on the record.

A Milestone Message:

"Trust is good, but a 458-page regulation is better."

Conclusion

The "Wild West" of AI in Europe is over. Compliance is the new innovation. Buckle up for a 2026 filled with audits, evaluations, and a lot of paperwork.