GitHub Repository

A zero-knowledge cryptographic steganography engine for AI dataset licensing. Built in Rust.

0 starsRust

Sigil – A zero-knowledge steganography vault for AI training data

Name: Sigil – A zero-knowledge steganography vault for AI training data
Availability: InStock
Author: nishalk

by nishalk·Mar 28, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●MidBold BetShip It

LSB steganography breaks under compression, making Nightshade a stronger choice for artists.

Strengths

•Open-source extractor allows third-party verification without installing the closed-source vault.
•Rust implementation ensures memory-safe pixel parsing for the extraction logic.
•Targets the specific friction point of unlicensed AI dataset procurement.

Weaknesses

•Closed-source vault undermines trust for a security tool claiming zero-knowledge.
•Standard image processing like JPEG compression or resizing strips the hidden payload.

Post Description

Hi HN,

I’ve been experimenting with ways to protect digital assets from automated AI web scrapers. I built Sigil, a local-first desktop application that embeds cryptographically signed ownership IDs into the LSB (Least Significant Bit) layer of images.

The desktop vault itself (Svelte/Tauri/SQLite) is closed-source to protect the HMAC-SHA256 signing architecture, but I have open-sourced the Rust extraction standard today.

The idea is to create an open protocol where AI data procurement teams can run the extractor against their scraped datasets. If it detects a payload, they know the asset is cryptographically locked and requires a license.

The extractor logic relies entirely on image and hex crates for memory-safe pixel parsing. I'd love any feedback on the extraction architecture or the LSB approach.

Extractor Repo: https://github.com/nishal21/Sigil-extractor Landing Page (Astro): https://nishal21.github.io/Sigil-extractor/