PulseAugur
EN
LIVE 21:59:42

New LLM watermarking techniques aim to protect IP and track usage

By PulseAugur Editorial · [4 sources] ·

Researchers have developed new methods for watermarking large language models (LLMs) to protect intellectual property and track usage. ArcMark, one new technique, embeds multiple bytes of information into text without altering the LLM's output distribution or perplexity. Another approach, SAFESEAL, uses key-conditioned sampling to preserve semantic fidelity and detect ownership, even against adversarial attacks. TextSeal, a third method, offers localized detection and can transfer its watermark signal through model distillation, making it effective against unauthorized use and replication. AI

IMPACT These watermarking advancements could enable better tracking of LLM-generated content and protect against unauthorized use and distillation.

RANK_REASON The cluster contains multiple academic papers detailing new research on LLM watermarking techniques.

Read on arXiv cs.CL →

AI-generated summary · Google Gemini · from 4 sources. How we write summaries →

COVERAGE [4]

  1. arXiv cs.AI TIER_1 English(EN) · Atefeh Gilani, Sajani Vithana, Carol Xuan Long, Oliver Kosut, Lalitha Sankar, Flavio P. Calmon ·

    ArcMark: Distortion-Free Multi-Byte LLM Watermark via Optimal Transport

    arXiv:2602.07235v2 Announce Type: replace-cross Abstract: Watermarking is an important tool for promoting the responsible use of large language models (LLMs). Existing watermarks insert a signal into generated tokens that either flags LLM-generated text (zero-bit watermarking) or…

  2. arXiv cs.CL TIER_1 English(EN) · Kieu Dang, Phung Lai, NhatHai Phan, Yelong Shen, Ruoming Jin ·

    Robust LLM Watermarking with Minimal Semantic Distortion for IP Protection

    arXiv:2605.23175v1 Announce Type: cross Abstract: Proprietary large language models (LLMs) face risks of intellectual property (IP) violation, as adversaries can replicate an LLM by collecting input-output pairs to train a surrogate model, causing financial setbacks. Watermarks o…

  3. arXiv cs.CL TIER_1 English(EN) · Tom Sander, Hongyan Chang, Tom\'a\v{s} Sou\v{c}ek, Tuan Tran, Valeriu Lacatusu, Sylvestre-Alvise Rebuffi, Alexandre Mourachko, Surya Parimi, Christophe Ropers, Rashel Moritz, Vanessa Stark, Hady Elsahar, Pierre Fernandez ·

    TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection

    arXiv:2605.12456v2 Announce Type: replace-cross Abstract: We introduce TextSeal, a state-of-the-art watermark for large language models. Building on Gumbel-max sampling, TextSeal introduces dual-key generation to restore output diversity, along with entropy-weighted scoring and m…

  4. arXiv cs.CL TIER_1 English(EN) · Ruoming Jin ·

    Robust LLM Watermarking with Minimal Semantic Distortion for IP Protection

    Proprietary large language models (LLMs) face risks of intellectual property (IP) violation, as adversaries can replicate an LLM by collecting input-output pairs to train a surrogate model, causing financial setbacks. Watermarks offer a promising defense to verify ownership, but …

RELATED ENTITIES

RELATED TOPICS