Paraphrasing Evades Detectors of AI-Generated Text, but Retrieval Is an Effective Defense

Krishna, K., Song, Y., Karpinska, M., Wieting, J., & Iyyer, M. (2023). Paraphrasing Evades Detectors of AI-Generated Text, but Retrieval Is an Effective Defense. Advances in Neural Information Processing Systems.

Type

Conference paper

Intellectual base

Primary research

Year

2023

Link

https://arxiv.org/abs/2303.13408

Cited from

security_intelligence

Cited by

1 citation across 1 artifact.

Each citation links to the sentence it supports in the citing article.

Primes

• Adversarial Boundary Navigation
  • Content and integrity evasion. Coded language and substitution to evade keyword filters, and obfuscation tuned to plagiarism or AI-detection classifiers.

    This sourceShows that paraphrasing (DIPPER) drops AI-text-detector accuracy toward chance while preserving semantics (DetectGPT from 70.3% to 4.6% at fixed FPR) — obfuscation tuned to a detection classifier, the content/integrity-evasion case.

Verification

This reference passed the adversarial substantiation pipeline: it was checked to exist and to support the claim it is attached to. See how references were verified.

Registry ID ref:fceccc84fa1e · see in the full table