New CARTS method uses Llama 3 to encode text payloads via rank transcoding

Researchers have developed a new method called Contextual Autoregressive Rank Transcoding Steganography (CARTS) that leverages autoregressive language models to embed text payloads within stegotext of identical token length. This technique preserves per-position rank information across different contexts. The paper provides the first formal security analysis of CARTS, demonstrating its correctness under deterministic model assumptions and introducing a rank-coordinate representation where keys act as bijections. An empirical study using Llama 3 8B confirmed exact payload recovery and suggested resistance to studied attack vectors. AI

IMPACT This research opens a formally grounded avenue for using language models in cryptography and privacy-preserving communication.

RANK_REASON The cluster contains an academic paper detailing a new method for text encoding using language models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Calgacus
• Contextual Autoregressive Rank Transcoding Steganography
• Llama 3 8B
• Norelli et al.

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →