Observed arrival · 2026-09-23
Garfield-Naruto Encoder hides messages in LLM-generated text
A text-steganography tool encodes a secret message into generated text, then offers a decoder to recover it with the same key.
- For
- People experimenting with text steganography
- Worth noticing
- Its security section says the V0 prototype is not recommended for the most security-critical applications.
Field notes
The described pipeline compresses the message with unishox2, then uses a key-based partition of characters to map generated text to bits or skips. At each inference step, token choices that would encode the wrong bit are removed; the page says hyperparameters adjust the balance between encoding density and indistinguishability. It also distinguishes its weaker security target from probability-preserving encoding and warns that the V0 prototype is not suitable for the most security-critical applications.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue