This research critically evaluates the reliability of AI-generated text detectors and reveals significant performance drops when applied to real-world content.

• Detection models achieving 99.9% accuracy in controlled tests often fail dramatically in practical applications
• Most benchmark datasets contain obvious artifacts that make detection artificially easy
• Researchers found major quality issues in 24 popular datasets used to train and test AI detectors
• Simple text modifications can significantly reduce detection accuracy

For security professionals, this highlights critical vulnerabilities in our ability to authenticate content sources and protect against AI-generated misinformation at scale.

Are AI Detectors Good Enough? A Survey on Quality of Datasets With Machine-Generated Texts