The SynthID Mirage: How Diffusion Models Are Dismantling Google’s AI Watermark
New research reveals diffusion-based attacks can strip Google’s SynthID watermark with 79% success, exposing fundamental flaws in AI content provenance systems and sparking a legal and ethical firestorm.
