As AI-generated content floods the web, the instinctive response from platforms has been to promise “detection.”

Watermarks. Signals. Provenance layers. Tools like Google’s SynthID are often framed as the beginning of a safer, more trustworthy internet.

They’re not. And they never will be—at least not at global scale.

The reason isn’t technical incompetence. It’s structural, economic, and deeply aligned with how the modern web actually works.

The Core Problem: Detection Requires Universal Cooperation

For AI detection to work at scale, every major AI system would need to participate in a shared standard for watermarking, signaling, and verification. That includes:

• All frontier AI labs
• All open-source model forks
• All future models that don’t yet exist

That level of coordination has never happened in tech history. Not for tracking cookies. Not for spam. Not for DRM. Not for privacy. And it won’t happen here either.

Detection Only Works Inside Walled Gardens

Tools like SynthID work best inside controlled ecosystems—Google Search, YouTube, Google Cloud. That’s not a bug; it’s the design constraint.

Once AI-generated content leaves the platform that created it, all bets are off:

• Screenshots strip metadata
• Re-uploads recompress files