← Back to archive

Watermark Models vs Rephrase Models

19 August 2026 · 45 views · 1 likes · 0 reposts · 138 blog views

AI Startups automation coding product

As you all know, Anthropic is going to watermark text, and of course that sparked a lot of thinking about how it could be done. One idea that immediately came to me was to hide the text using Unicode. I quickly built a prototype (see the link below), but later realized it’s probably too easy -though it might survive Copy & Paste, it’s really just a first layer. The second layer would be statistical or sampling-based watermarks, which I think could be solved with rephrasing.

I can see more and more startups getting into this, and building their own "rephrase" models that can be fine-tuned to match the company’s/user's voice.

We'll end up with watermark models fighting watermark-removal models and every generation of one creating training data for the other.

the tool: desunit.com/blog/projects/…

View on X

Watermark Models vs Rephrase Models

Originally posted on X

PS: I'm pretty active on 𝕏 if you'd like to follow along

Related posts

Archive

2026

2025

2024

2018