tech
April 14, 2026
Has Google’s AI watermarking system been reverse-engineered?
Posts from this topic will be added to your daily email digest and your homepage feed.
TL;DR
- A developer claims to have reverse-engineered Google DeepMind's AI watermarking system, SynthID.
- The developer, Aloshdenny, open-sourced their method on GitHub, describing it as achievable with 200 generated images and signal processing.
- Aloshdenny's technique involves averaging patterns from AI-generated images to expose and manipulate watermark signals.
- Google disputes Aloshdenny's claims, asserting that their SynthID system is robust and cannot be systematically removed by the described tool.
- The effectiveness of Aloshdenny's project has not been independently verified, and Google maintains that their watermarking technology is effective.
Continue reading the original article