Has Google’s AI watermarking system been reverse-engineered?

A software developer using the username Aloshdenny claims to have reverse-engineered Google DeepMind's SynthID watermarking system, which is designed to identify AI-generated images and protect against misuse. According to the developer's assertions, the system can be bypassed to remove watermarks from AI-generated content or fraudulently insert them into non-AI images. This claim directly contradicts Google's position, which denies that such reverse-engineering has occurred or that the system is vulnerable in the manner described.

The alleged vulnerability raises significant concerns about the effectiveness of current AI watermarking technologies in combating misinformation and unauthorized use of generated content. If legitimate, such a breach would undermine efforts to track and verify the authenticity of AI-generated images, a critical issue as synthetic media becomes increasingly sophisticated and prevalent online.

The dispute highlights the ongoing security challenges in AI systems and the gap between theoretical protections and real-world vulnerabilities. The outcome of this claim will likely influence confidence in watermarking as a reliable method for AI content verification and may prompt deeper scrutiny of similar protection mechanisms used by other technology companies.