# Every Bit, Everywhere, All at Once: A Binomial Multibit LLM Watermark Source: [Every Bit, Everywhere, All at Once: A Binomial Multibit LLM Watermark]() Collection index: [llms.txt]() Data updated: 2026-05-14T05:00:05.131Z Authors: Thibaud Gloaguen, Robin Staab, Mark Vero, Martin Vechev Published: 2026-05-12T07:14:45.000Z ## Abstract With LLM watermarking already being deployed commercially, practical applications increasingly require multibit watermarks that encode more complex payloads, such as user IDs or timestamps, into the generated text. In this work, we propose a fundamentally new approach for multibit watermarking: introducing binomial encoding to directly encode every bit of the payload at every token position. We complement our approach with a stateful encoder that during generation dynamically redirects encoding pressure toward underencoded bits. Our evaluation against 8 baselines on up to 64-bit payloads shows that our scheme achieves superior message accuracy and robustness, with the gap to baseline methods widening in more relevant settings (i.e., large payloads and low-distortion regimes). At the same time, we challenge prior works' evaluation metrics, highlighting their lack of practical insights, and introduce per-bit confidence scoring as a practically relevant metric for evaluating multibit LLM watermarks. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Synonym Substitution Robustness on ELI5 prompts 32-bit payload Llama3.1-8B (test)](): Synonym Substitution Robustness; ELI5 prompts 32-bit payload Llama3.1-8B (test); Bit Accuracy (10% Substitution): 70.7. - [Word Deletion Robustness on ELI5 prompts 32-bit payload Llama3.1-8B (test)](): Word Deletion Robustness; ELI5 prompts 32-bit payload Llama3.1-8B (test); Bit Accuracy (10% Deletion): 70.7. - [Distortion-Free Watermark Evaluation on ELI5 32-bit bitstrings LLAMA3.1-8B](): Distortion-Free Watermark Evaluation; ELI5 32-bit bitstrings LLAMA3.1-8B; Message Accuracy: 0.3. - [Paraphrasing Robustness on ELI5 prompts 32-bit payload Llama3.1-8B (test)](): Paraphrasing Robustness; ELI5 prompts 32-bit payload Llama3.1-8B (test); Bit Accuracy: 50.9. - [Distortion-Free Watermark Evaluation on ELI5 16-bit bitstrings LLAMA3.1-8B](): Distortion-Free Watermark Evaluation; ELI5 16-bit bitstrings LLAMA3.1-8B; Message Accuracy: 15.8. - [Distortion-Free Watermark Evaluation on ELI5 64-bit bitstrings LLAMA3.1-8B](): Distortion-Free Watermark Evaluation; ELI5 64-bit bitstrings LLAMA3.1-8B; Message Accuracy: 0. ## Original sources and implementation [Original paper]() [PDF]()