concept

watermarking

Also known as: watermarks

Facts (10)

Sources

A Survey on the Theory and Mechanism of Large Language Models arxiv.org arXiv Mar 12, 2026 9 facts

claimAI-generated text detection tools, such as watermarking, are critical for identifying machine-generated content and ensuring accountability.

referenceHe et al. (2024a) introduced a unified theoretical framework for watermarking Large Language Models that jointly optimizes the watermarking scheme and the detector, revealing a fundamental trade-off between watermark detectability (Type-II error) and text distortion.

referenceThe paper 'Undetectable watermarks for language models' was published in The Thirty Seventh Annual Conference on Learning Theory, pp. 1125–1139.

referenceThe paper 'Robust detection of watermarks for large language models under human edits' was published in the Journal of the Royal Statistical Society Series B: Statistical Methodology.

referenceThe paper 'A statistical framework of watermarks for large language models: pivot, detection efficiency and optimal rules' was published in The Annals of Statistics 53 (1), pp. 322–351.

referenceThe paper 'Provably robust watermarks for open-source language models' is an arXiv preprint (arXiv:2410.18861) cited in the context of language model security.

claimWatermarking allows the output of proprietary Large Language Models to be algorithmically identified as synthetic with negligible impact on text quality.

referenceThe paper 'A watermark for large language models' proposes a method for watermarking large language models.

claimChrist et al. (2024a) proved that watermarks in Large Language Models are unremovable under the assumption of adversary uncertainty about the high-quality text distribution, establishing a trade-off between quality degradation and watermark removal.

Engineering biology applications for environmental solutions - Nature nature.com Nature Apr 14, 2025 1 fact

claimWang and Zhang recommend the use of genomic barcodes or watermarks as complementary strategies to biocontainment for tracking engineered biological assets.