Hugging Face Daily Papers · · 3 min read

PASA: A Principled Embedding-Space Watermarking Approach for LLM-Generated Text under Semantic-Invariant Attacks

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Welcome</p>\n","updatedAt":"2026-05-13T04:28:02.429Z","author":{"_id":"67a4a26d5e65aa63c6d30e68","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/67a4a26d5e65aa63c6d30e68/GtodlJGw-_IL2DTXQTucz.jpeg","fullname":"Sicheng Feng","name":"FSCCS","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":12,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.3305741548538208},"editors":["FSCCS"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/67a4a26d5e65aa63c6d30e68/GtodlJGw-_IL2DTXQTucz.jpeg"],"reactions":[],"isReport":false}},{"id":"6a0482ad255af1730ef0211c","author":{"_id":"661ab1f1fa3b144a381fa454","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/661ab1f1fa3b144a381fa454/IlpZBb9NCjo7ntFwMIH53.png","fullname":"Urro","name":"urroxyz","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false},"createdAt":"2026-05-13T13:54:53.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Wow! Genius!\n\nText watermarking is important. I hope that stable research on it gets adopted so that communities can fight AI-generated spam.","html":"<p>Wow! Genius!</p>\n<p>Text watermarking is important. I hope that stable research on it gets adopted so that communities can fight AI-generated spam.</p>\n","updatedAt":"2026-05-13T13:54:53.237Z","author":{"_id":"661ab1f1fa3b144a381fa454","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/661ab1f1fa3b144a381fa454/IlpZBb9NCjo7ntFwMIH53.png","fullname":"Urro","name":"urroxyz","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9107217192649841},"editors":["urroxyz"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/661ab1f1fa3b144a381fa454/IlpZBb9NCjo7ntFwMIH53.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.10977","authors":[{"_id":"6a03f8f186b054ce2fa40eea","user":{"_id":"66def1e3ba8b9dac859dbd64","avatarUrl":"/avatars/84797ac61013046db3a495d5033f9d32.svg","isPro":false,"fullname":"Zhenxin Ai","user":"kunkk","type":"user","name":"kunkk"},"name":"Zhenxin Ai","status":"claimed_verified","statusLastChangedAt":"2026-05-13T07:43:53.369Z","hidden":false},{"_id":"6a03f8f186b054ce2fa40eeb","name":"Haiyun He","hidden":false}],"publishedAt":"2026-05-09T00:00:00.000Z","submittedOnDailyAt":"2026-05-13T00:00:00.000Z","title":"PASA: A Principled Embedding-Space Watermarking Approach for LLM-Generated Text under Semantic-Invariant Attacks","submittedOnDailyBy":{"_id":"67a4a26d5e65aa63c6d30e68","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/67a4a26d5e65aa63c6d30e68/GtodlJGw-_IL2DTXQTucz.jpeg","isPro":false,"fullname":"Sicheng Feng","user":"FSCCS","type":"user","name":"FSCCS"},"summary":"Watermarking for large language models (LLMs) is a promising approach for detecting LLM-generated text and enabling responsible deployment. However, existing watermarking methods are often vulnerable to semantic-invariant attacks, such as paraphrasing. We propose PASA, a principled, robust, and distortion-free watermarking algorithm that embeds and detects a watermark at the semantic level. PASA operates on semantic clusters in a latent embedding space and constructs a distributional dependency between token and auxiliary sequences via shared randomness synchronized by a secret key and semantic history. This design is grounded in our theoretical framework that characterizes a jointly optimal embedding-detection pair, achieving the fundamental trade-offs among detection accuracy, robustness, and distortion. Evaluations across multiple LLMs and semantic-invariant attacks demonstrate that PASA remains robust even under strong paraphrasing attacks while preserving high text quality, outperforming standard vocabulary-space baselines. Ablation studies further validate the effectiveness of our hyperparameter choices. Webpage: https://ai-kunkun.github.io/PASA_page/.","upvotes":9,"discussionId":"6a03f8f186b054ce2fa40eec","projectPage":"https://ai-kunkun.github.io/PASA_page/","githubRepo":"https://github.com/ai-kunkun/PASA","githubRepoAddedBy":"user","ai_summary":"PASA is a robust watermarking algorithm for large language models that operates at the semantic level using latent embedding spaces and shared randomness for secure text detection.","ai_keywords":["watermarking","large language models","semantic level","latent embedding space","shared randomness","secret key","semantic history","detection accuracy","robustness","distortion"],"githubStars":17,"organization":{"_id":"65ad19cac14c3cf579ad9b68","name":"HKUSTGZ","fullname":"HKUSTGZ","avatar":"https://www.gravatar.com/avatar/70f662accc42fce026543de48099c80f?d=retro&size=100"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"66def1e3ba8b9dac859dbd64","avatarUrl":"/avatars/84797ac61013046db3a495d5033f9d32.svg","isPro":false,"fullname":"Zhenxin Ai","user":"kunkk","type":"user"},{"_id":"67a4a26d5e65aa63c6d30e68","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/67a4a26d5e65aa63c6d30e68/GtodlJGw-_IL2DTXQTucz.jpeg","isPro":false,"fullname":"Sicheng Feng","user":"FSCCS","type":"user"},{"_id":"6919d7b335f761536e60fdf4","avatarUrl":"/avatars/e4a58a666da08da56ab09cef4a3e75a5.svg","isPro":false,"fullname":"Rowan Ellis","user":"Simon203","type":"user"},{"_id":"66aa39349238d9c3a1c7f9dc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66aa39349238d9c3a1c7f9dc/mj6r7uxEYXM502x296UMf.jpeg","isPro":false,"fullname":"Xin Jin","user":"Xin1118","type":"user"},{"_id":"655469586bc4180700cf7a34","avatarUrl":"/avatars/252392d0c45783d8f149feac7a6215ec.svg","isPro":false,"fullname":"Kejia Zhang","user":"KejiaRobust","type":"user"},{"_id":"67d5848f179ad2756600eca3","avatarUrl":"/avatars/158168a753271b6e024e1fbdf52c9e73.svg","isPro":false,"fullname":"Junhan ZHU","user":"Alrightlone","type":"user"},{"_id":"681086965ac594b801241cc9","avatarUrl":"/avatars/e4d0440141010deec595846185c98714.svg","isPro":false,"fullname":"Jianlong He","user":"elaxEgan","type":"user"},{"_id":"661ab1f1fa3b144a381fa454","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/661ab1f1fa3b144a381fa454/IlpZBb9NCjo7ntFwMIH53.png","isPro":true,"fullname":"Urro","user":"urroxyz","type":"user"},{"_id":"6a0464d0abaa37d1f4506758","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a0464d0abaa37d1f4506758/qZ6kK8_hdYai26VDiK6GM.jpeg","isPro":false,"fullname":"kuzcaaxdwn","user":"kuzcaaxdwn","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"65ad19cac14c3cf579ad9b68","name":"HKUSTGZ","fullname":"HKUSTGZ","avatar":"https://www.gravatar.com/avatar/70f662accc42fce026543de48099c80f?d=retro&size=100"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.10977.md"}">
Papers
arxiv:2605.10977

PASA: A Principled Embedding-Space Watermarking Approach for LLM-Generated Text under Semantic-Invariant Attacks

Published on May 9
· Submitted by
Sicheng Feng
on May 13
Authors:

Abstract

PASA is a robust watermarking algorithm for large language models that operates at the semantic level using latent embedding spaces and shared randomness for secure text detection.

AI-generated summary

Watermarking for large language models (LLMs) is a promising approach for detecting LLM-generated text and enabling responsible deployment. However, existing watermarking methods are often vulnerable to semantic-invariant attacks, such as paraphrasing. We propose PASA, a principled, robust, and distortion-free watermarking algorithm that embeds and detects a watermark at the semantic level. PASA operates on semantic clusters in a latent embedding space and constructs a distributional dependency between token and auxiliary sequences via shared randomness synchronized by a secret key and semantic history. This design is grounded in our theoretical framework that characterizes a jointly optimal embedding-detection pair, achieving the fundamental trade-offs among detection accuracy, robustness, and distortion. Evaluations across multiple LLMs and semantic-invariant attacks demonstrate that PASA remains robust even under strong paraphrasing attacks while preserving high text quality, outperforming standard vocabulary-space baselines. Ablation studies further validate the effectiveness of our hyperparameter choices. Webpage: https://ai-kunkun.github.io/PASA_page/.

Community

Paper submitter about 17 hours ago

Welcome

Wow! Genius!

Text watermarking is important. I hope that stable research on it gets adopted so that communities can fight AI-generated spam.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images

· Sign up or log in to comment

Get this paper in your agent:

hf papers read 2605.10977
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2605.10977 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2605.10977 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2605.10977 in a Space README.md to link it from this page.

Collections including this paper 1

Discussion (0)

Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.

Sign in →

No comments yet. Sign in and be the first to say something.

More from Hugging Face Daily Papers