Google's SynthID survives the stress test — but will it stop misinformation?

ars technica tested the durability of google's synthid watermarking technology by subjecting ai-generated and edited images to extreme compression and resizing using a python script. the results showed that synthid remains detectable even after severe data loss, outperforming cryptographically secure metadata like c2pa which is easily stripped by simple actions like taking a screenshot. google deepmind scientists confirmed the system was designed to withstand attacks and image transformations, though specific technical details remain sparse. despite its resilience, the report notes that watermarks are not a silver bullet for misinformation, as they require universal adoption across openai, runway, and nvidia to be effective.