The short version
A library for hiding steganographic signals in language-model output.
From the HackerLinks archive
Innocuous
A library for hiding steganographic signals in language-model output.
At a glance:
First seen:2026-09-23
Last seen:2026-09-23
Times seen:1
Website:github.com
Why it caught our attention
Its author reported a working implementation closely matching Anthropic's watermark approach.
Where it surfaced on Hacker News
2026-09-23
“I actually wrote a library to do stego with LLM outputs last year and it turned out to be an almost exact implementation of the Anthropic watermark algo.”
stillsut · author context · author or maintainer
Direct HN comment
Spymarks, not Watermarks