AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Catch-22: On the Fundamental Tradeoff Between Detectability and Robustness in LLM Watermarking

Kuheli Pratihar, Debdeep Mukhopadhyay

ICML 2026 · Large language models › Language, safety and interpretability · approach: Text watermarking

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add DIPPER paraphrasing attack to HCW (bias-free)HCW · LFQA Llama-2-7BAUROC: .99 -> .91Hurt
use semantic PMark instead of bias-free HCW under DIPPERwatermark family · LFQA, DIPPERAUROC: .91 -> .94Improved
use semantic PMark instead of bias-free HCW under OPT-2.7Bwatermark family · LFQA, OPT-2.7BAUROC: .92 -> .95Improved
increase edit rate to back-translation (~0.42) for bias-free schemesbias-free watermarks · LFQA Llama-2-7BAUROC: .99 -> .55 to .60Hurt
use semantic PMark instead of HCW under back-translationwatermark family · LFQA, back-translationAUROC: .59 -> .85Improved
use hybrid selection instead of best single family under back-translationHybrid · LFQA, back-translationAUROC: .85 -> .86Improved
use hybrid instead of biased KGW (no attack)Hybrid · LFQA, no attackkeyless z-score: 30.1 -> −7.8Improved
use distribution-preserving CGW instead of KGW under DIPPERCGW · LFQA, DIPPERAUROC: .86 -> .50Hurt
use hybrid instead of KGW under DIPPERHybrid · LFQA, DIPPERTPR at 1% FPR: .64 -> .89Improved
use co-designed DAWA instead of Hybrid under DIPPERDAWA · LFQA, DIPPERAUROC: .94 -> .54Hurt
use conservative upper-confidence-bound selector (ablation)Hybrid selector · Appendix G sensitivityworst-case regretImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy