AI 101

Watermarking

Also known as: AI watermark · text watermark

Security

Watermarking embeds a subtle, detectable signal in generated content so you can verify it came from a specific model or process.

  • Text: biased sampling toward a word set.
  • Images: a pattern in frequency space.
  • Goal: provenance and (theoretically) abuse control.

The honest state: watermarking helps with provenance, but it's fragile — easy to strip or bypass. Treat it as one layer, not a guarantee.