TL;DR

A soft green-token watermark embedded during LLM sampling is detectable via a statistical hypothesis test with no model access required, published at ICML 2023.

Summary

Detecting AI-generated text without model access is the central challenge this paper solves. A random green list of tokens is selected before each generation step and softly promoted during sampling; a statistical hypothesis test on the proportion of green tokens then detects the watermark with interpretable p-values — no API access or model weights required. Tested on the OPT model family, the method produces negligible quality degradation while remaining robust to moderate text edits.

Key contributions

  1. Proposes a green-token watermarking scheme that biases sampling toward a secret random token subset without requiring model retraining.
  2. Introduces a statistical detection test based on green-token proportion that requires only the secret key, not model internals.
  3. Demonstrates negligible text quality degradation on OPT multi-billion parameter models.
  4. Establishes robustness bounds: the watermark survives moderate paraphrasing but can be defeated by targeted rewriting attacks.

When to cite

  1. When discussing methods for detecting AI-generated text without access to model weights or APIs.
  2. When arguing that watermarking and generation quality are not fundamentally in tension.
  3. When evaluating the robustness limits of LLM output provenance techniques against adversarial paraphrasing.
  4. When comparing approaches to LLM output attribution that rely on statistical rather than cryptographic guarantees.

LLM GPT On the Reliability of Watermarks for Large Language Models