Greece eyes prison terms for removing AI deepfake watermarks

Greece is stress-testing how far democracies are willing to go in criminalizing abuses of synthetic media. The underlying EU AI Act already requires deepfakes to be clearly labeled, but leaves enforcement to member states. By moving from fines to possible prison time for removing watermarks, Athens is signalling that it sees AI‑generated disinformation as closer to election tampering or hate speech than to ordinary copyright infringement.

For the broader AI race, this is part of a pattern: as generative models get better at mimicking reality, the regulatory perimeter keeps expanding from models themselves to the entire information stack—tooling, watermarks, platforms and end‑users. That creates real compliance overhead for anyone shipping open‑weight models or media tools into the EU, especially if other countries copy Greece’s approach. At the same time, aggressive penalties could push actors toward more robust watermarking standards and verification protocols, which in turn may force frontier labs to bake provenance and traceability more deeply into their training and inference pipelines.