Google’s AI Invisible Watermark Will Help Identify Generative Text and Video

Among Google’s slew of new AI models and tools announced today, the company is also expanding its AI content detection and watermarking technology to work across two new media.

Google’s DeepMind CEO Demis Hassabis took the stage for the first time at the Google I/O developer conference on Tuesday to talk not only about the team’s new AI tools like the Veo video generator, but also about the new updated SynthID watermark printing. system. It can now mark up digitally generated videos as well as AI-generated texts.

SynthID was announced last August and began as a tool to print AI images in a way that humans cannot decipher visually – but can be detected by the system. The approach is different from other aspiring watermarking protocol standards like C2PA, which add cryptographic metadata to AI-generated content.

Google also allowed SynthID to inject inaudible watermarks into AI-generated music made using DeepMind’s Lyria model. SynthID is just one of several AI safeguards being developed to combat misuse of the technology, safeguards that the Biden administration is directing federal agencies to create guidelines for.