AI-generated media is becoming easier to create and harder to verify. That makes watermarking, provenance, and labeling some of the most important technology problems in generative AI.
Google SynthID helping identify a viral fake image shows why durable watermarks matter. Google’s rollback of an AI image feature inside Google Earth shows the other side of the problem: trusted interfaces can make synthetic content look more credible than it is.
What Needs To Improve
- AI images and videos need durable provenance signals.
- Platforms need labels that survive reposting and screenshots.
- Verification tools must be easy for journalists and users to access.
- Trusted products like maps need stricter rules for AI-generated visuals.
Generative media is not automatically harmful. The problem is unlabeled synthetic content presented as evidence.
Should major AI media tools be required to watermark generated content by default?