Featured story
Deepfake
Google's SynthID Takes on AI Deepfakes
Google is strengthening its fight against AI-generated misinformation through SynthID Detector, a verification platform introduced at Google I/O 2025 and expanded in 2026 to identify AI-generated text, images, videos and audio. The platform detects invisible digital watermarks embedded in content created using Google's AI models, including Gemini, Imagen, Veo and Lyria. Users can upload files to identify watermarked content, including specific segments within audio recordings. Unlike conventional metadata-based identification, SynthID embeds watermarks directly into pixels, audio waveforms and text-generation patterns. Two coordinated AI models handle watermark insertion and detection, making identification more resilient to compression, cropping and other modifications. Google's advantage lies in controlling the underlying generative AI ecosystem. By embedding watermarks during content creation, the company can establish provenance signals rather than relying entirely on forensic analysis after content circulates online. Google has reportedly extended SynthID verification into Search through Lens, AI Mode and Circle to Search, with broader ecosystem integration underway. Its scale and DeepMind's research capabilities strengthen adoption across content-generation platforms. However, SynthID has a fundamental limitation: it primarily identifies content carrying compatible watermarks. Unmarked AI-generated material, manipulated recordings and sophisticated deepfakes from unsupported systems may escape detection. Watermarks can also face deliberate removal attempts. Consequently, watermarking must complement independent deepfake detection, biometric liveness verification, behavioural intelligence and forensic analysis. Combined with C2PA Content Credentials, these technologies can strengthen digital provenance and trust. The future of AI security requires both verifying where content originated and independently examining whether its contents have been manipulated.
Read full story
