PulseAugur
EN
LIVE 03:38:37

New watermarking technique attributes code to LLMs like GPT-4.1 and Llama 4

By PulseAugur Editorial · [2 sources] ·

Researchers have developed a novel multi-channel spread-spectrum code watermarking technique that can attribute code to its originating large language model. This post-hoc, training-free method offers a 24-bit payload, significantly more than previous methods, and provides formal robustness guarantees against various attacks. Tested on Python files generated by GPT-4.1 and Llama 4, the watermark achieved 100% detection accuracy and maintained high accuracy even under significant corruption and transformation attacks. AI

IMPACT Enables better tracking of AI-generated code for provenance, licensing, and accountability.

RANK_REASON The cluster describes a novel research paper detailing a new watermarking technique for code generated by LLMs.

Read on arXiv cs.LG →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

New watermarking technique attributes code to LLMs like GPT-4.1 and Llama 4

COVERAGE [2]

  1. arXiv cs.LG TIER_1 English(EN) · Soohyeon Choi, Debin Gao, Yue Duan ·

    Multi-Channel Spread-Spectrum Code Watermarking

    arXiv:2607.06009v1 Announce Type: cross Abstract: Attributing code to the large language model that produced it is essential for provenance, licensing, and misuse accountability, yet no deployed watermark meets this need. Generation-time schemes require access to the producing mo…

  2. arXiv cs.LG TIER_1 English(EN) · Yue Duan ·

    Multi-Channel Spread-Spectrum Code Watermarking

    Attributing code to the large language model that produced it is essential for provenance, licensing, and misuse accountability, yet no deployed watermark meets this need. Generation-time schemes require access to the producing model and cannot be applied to third-party code, whi…

RELATED ENTITIES

RELATED TOPICS