Google’s SynthID is hard to break, but it still won’t fix AI misinformation
davidstutz92 · x · 2026-07-29
Ars Technica reports that Google’s SynthID watermark is proving difficult to remove, but that alone will not solve the misinformation problem.
The article notes the scale of AI-generated media: humanity took until 1975 to make 1.5 billion images after the camera was invented, while generative AI matched that volume in just 18 months. Google also says its tools were used to create more than 100 billion AI images and videos in only a couple of years.
Google is now expanding SynthID through partnerships with OpenAI, Runway, Nvidia, and others. The piece argues that invisible watermarks are only one part of a broader provenance problem; metadata standards and other labeling systems are also needed if users are going to reliably tell what is real online.
More from Safety
- Anthropic grants METR inside access to audit its AI models and safety practices — The Decoder · 2026-10-16
- Agent actions need checks outside the model: NVIDIA demos lean on guardrails amid wave of real agent incidents — Creamy-And-Crowded · 2026-10-16
- Vitalik builds "paranoid mode" agent wrapper with Tor, zkAPI and local prompt rewriting — DavideCrapis · 2026-10-16
- Security team demos Signal 1-click exploit technique at Hexacon Paris — moyix · 2026-10-16
- China moves to impose registration and licensing on frontier AI model weights and chip designs — deanwball · 2026-10-16
- Same agent task, two auth models: shared admin login wipes 1M audit rows, per-agent identity gets refused — Prompt Engineering · 2026-10-16