Google’s SynthID watermark is hard to break, but it doesn’t solve AI misinformation
Summary
Google's SynthID watermarking aims to label AI-generated content, but a test by Ars Technica shows it survives many transformations yet can be degraded or bypassed by cropping and heavy compression. The piece argues SynthID is not a silver bullet for AI misinformation, highlighting interoperability gaps with other detectors (like OpenAI's or Meta's watermarks) and the broader need for cryptographic provenance data (C2PA) and verifiable metadata. It also notes access limits and evolving security considerations as AI content proliferates.