AI Tool Conversation Steganography Hides Secrets in Soup Recipes

Developer Nethical published the open-source tool Conversation Steganography on GitHub, written in Go under the GPL v3.0 license. The software combines local artificial intelligence with arithmetic coding to hide confidential messages inside ordinary chat text, making secret exchanges look like trivial everyday talk.

A simple instruction like meeting at a coffee shop at 3 PM converts into a casual paragraph about fresh basil soup. The recipient's local neural network decrypts the text back into its original form using a shared secret phrase and matching mathematical parameters, leaving surveillance systems completely oblivious to the real payload.

The underlying security framework relies on AES-SIV encryption and PBKDF2-HMAC-SHA-256 key derivation with 600,000 iterations. Every message is cryptographically chained to its predecessor, running entirely offline on light models such as Llama 3 4b or ancient GPT-2 without cloud telemetry or external API calls.

Using the system requires absolute operational precision, as autocorrect, smart quotes, or out-of-order text insertion destroy the decryption process. Both participants must run the exact same model build, settings, and passphrase, copying generated cover text byte for byte.

Digital privacy has officially devolved into pretending to care about soup while running heavy cryptographic algorithms locally. When paranoia requires this much administrative effort, normal human conversations might soon become completely obsolete.

Source: GitHub