DART introduces an innovative method for detecting AI-generated texts (AIGT) by analyzing semantic structures rather than relying on probabilistic features.

Key Advances:

• Leverages Abstract Meaning Representation (AMR) of rephrased text to identify AI-generated content
• Addresses challenges in detecting texts from black-box LLMs
• Tests detection in multi-candidate settings that better reflect real-world scenarios
• Improves security by identifying potential misinformation or fraudulent content

This research matters for security professionals as detection of AI-generated content becomes critical for combating misinformation, verifying content authenticity, and maintaining digital security standards.

DART: An AIGT Detector using AMR of Rephrased Text