DGIST develops innovative text watermarking technology with a 2% false positive rate, detecting AI-generated text while safeguarding human writing

Posted on July 13, 2026
By Google News
0 Views
1 min read

In brief

South Korean researchers at DGIST have unveiled BREW, a groundbreaking watermarking system designed to identify artificially generated text with remarkable accuracy. The technology employs a block-based verification methodology that analyzes segmented text portions independently, maintaining exceptional precision even when content undergoes paraphrasing or word substitution. With only a 2% false positive rate on human-authored material, this innovation represents a significant advancement in digital authenticity verification and misinformation prevention.

Summary auto-generated by AI from the original publisher's content. Editorial standards.

DGIST develops innovative text watermarking technology with a 2% false positive rate, detecting AI-generated text while safeguarding human writing

- Prof. Young-Sik Kim's research team develops BREW, a block-based verification technique that independently verifies segmented text blocks - Detects AI-generated text even after word substitutions and paraphrasing, reducing the false positive rate for human-written text to 2% - A core AI forensic technology for preventing fake news, accepted to the prestigious International Conference on Machine Learning (ICML)