EU AI Act's August 2026 transparency requirement faces technical impossibility with current generative AI systems, creating compliance gridlock.

Article 50 II of the EU AI Act mandates dual transparency (human-readable and machine-readable labels) for AI-generated content starting August 2026

Fact-checking systems cannot comply due to non-deterministic LLM outputs and iterative workflows that make provenance tracking infeasible

Synthetic data generation faces a paradox: watermarks visible to humans risk being learned as spurious features during model training, rendering compliance self-defeating

The assistive-function exemption doesn't apply to fact-checking systems since they actively assign truth values rather than merely support editorial decisions