An open academic framework for stealth data transmission, prepositional text encoding, and neural carrier verification built for high-security plain text communication.
Get StartedGenerates coherent email threads with embedded prepositional payloads, allowing for secure message transmission within natural text.
Generates narrative text with embedded prepositional payloads, allowing for secure message transmission within natural text.
Generates academic essays with embedded prepositional payloads, allowing for secure message transmission within natural text.
Two clear views of how a message is hidden in natural text and recovered from it.
The secret message is converted into hexadecimal character codes. Each hex digit can then be represented by a preposition.
Each hexadecimal digit maps to one preposition from a password-derived ordering. The prepositions are assembled in sequence as the hidden payload.
A language model writes a coherent passage around the required prepositions, concealing them in ordinary-looking text.
Decoding reverses the encoding steps. Provide the generated carrier text and the same password used to encode it; the decoder extracts the prepositions, maps them back to hexadecimal, and reconstructs the original message.
The decoder scans the carrier text and keeps supported prepositions in the order they appear, ignoring the surrounding cover words.
The password recreates the preposition mapping used during encoding. Each extracted preposition becomes its hexadecimal digit; every two digits form one character code.
The hexadecimal character codes are converted back into text and returned as the decoded payload.
Use the same password for both directions. If no password was set, decode with the default key.
Out of all the test runs across different message lengths, the decoder extracted the original secret message with 100% perfection nearly 40% of the time (with shorter messages performing as high as 63.3%)
For every single character in your secret message (e.g., the letter "A"), the LLM generates roughly 114.5 characters of cover text to blend the preposition-encoded payload naturally into readable prose.
On average, it takes about 30.5 seconds from the moment a trial is initiated for the model to craft and output the full encoded cover text.
Author: Vaishnavi Adidam - IEEE
You are navigating to the interactive module interface. In production, this routes directly to the operational steganography workspace.
