llm-watermarking

[USENIX Security'24] REMARK-LLM: A robust and efficient watermarking framework for generative large language models

Test what AI watermark and provenance evidence exists, whether it verifies, and what survives publishing.

An application that uses a multistage pipeline to remove any watermarks from AI generated text

Open-source AI text watermark remover for Unicode artifacts, with detector-scoped evaluation of published statistical LLM watermark schemes.

[ICML'25] An End-to-End Model for Logits Based Large Language Models Watermarking

Geometry of meaning-preserving rewriting paths and the erosion of statistical text watermarks.

spab is a watermarking tool for text/code/md files which can embed a hash, string, or message.

Streamlit app exploring LLM text watermarking, token-level and substring-based detection

FP-18: LLM Watermark Robustness Under Adversarial Paraphrasing

Capstone Project 2025 (Yonsei Univ.)

Detect and remove hidden Unicode text watermarks locally, then verify LLM watermark mitigations with evidence-based tools and a CLI.

Inspect AI provenance evidence, verify watermarks, and assess what publishing workflows destroy before relying on any single authenticity signal.

Turkish LLM watermarking: detector calibration, scheme fragility, and the TR-WM-EVAL benchmark. Research fork of MarkLLM.