Google's AI Watermarking System Under Scrutiny
April 14, 2026 at 13:53
0
✦ AI Summary
- Developer claims to have reverse-engineered SynthID
- Process involved basic tools and 200 AI-generated images
- Google disputes the validity of these claims
A software developer has announced the potential reverse-engineering of Google DeepMind's SynthID system, which is designed for AI watermarking. This assertion involves the ability to remove watermarks from generated images or embed them into different works.
The individual, known as Aloshdenny, shared their findings on GitHub, revealing that the process required merely 200 images produced by Gemini, some signal processing techniques, and an abundance of time. Interestingly, they claimed that no advanced neural networks or privileged access were utilized.
Google has responded by rejecting these assertions, maintaining that the developer's claims do not hold true.
Share: