92
/ 100
True
Daegu South Korea
The website reports on DGIST's development of BREW, a text watermarking technology with a 2% false positive rate, capable of detecting AI-generated text even after modifications. The research is accepted for presentation at ICML 2026.
Infact verdict: True (92/100).
The claims regarding DGIST's development of BREW, a text watermarking technology, are well-supported by multiple credible sources. The technology's capabilities, including a 2% false positive rate and a 96.5% detection rate even with word substitutions, are corroborated by evidence from Asia Research News and other academic sources. Additionally, the acceptance of the research findings at ICML 2026 is confirmed by the conference's official blog. Overall, the evidence strongly supports the accuracy of these claims.
How is this score determined? →Individual claims
92
True
Technology
DGIST develops innovative text watermarking technology with a 2% false positive rate.
The evidence from Asia Research News confirms that DGIST has developed a text watermarking technology with a 2% false positive rate. This is corroborated by multiple sources discussing the technology's capabilities and limitations, such as its struggle against paraphrasing and plagiarism. The evidence is consistent and comes from credible sources, supporting the claim's accuracy.
Fact Check Score
None
Fact Check Weight
0
Web Consensus Score
90
Web Consensus Weight
40
Source Quality Score
80
Source Quality Weight
20
Llm Reasoning Score
85
Llm Reasoning Weight
40
Llm Reasoning Score Raw
None
Weighted Total
92
Evidence Summary
Multiple web sources confirm DGIST's development of watermarking technology with a 2% false positive rate.
89
True
Technology
BREW detects AI-generated text even after word substitutions and paraphrasing.
The claim is supported by evidence from Asia Research News and arXiv, which describe BREW's ability to detect AI-generated text even after modifications like word substitutions and paraphrasing. The technology uses advanced techniques to maintain a low false positive rate, which is corroborated by multiple sources.
Fact Check Score
None
Fact Check Weight
0
Web Consensus Score
85
Web Consensus Weight
40
Source Quality Score
75
Source Quality Weight
20
Llm Reasoning Score
80
Llm Reasoning Weight
40
Llm Reasoning Score Raw
None
Weighted Total
89
Evidence Summary
Multiple sources confirm BREW's detection capabilities even after text modifications.
92
True
Technology
BREW maintained a high detection rate of 96.5% even when 10% of the words in AI-generated text were replaced with synonyms.
The claim is supported by evidence from Asia Research News, which confirms that BREW maintained a 96.5% detection rate even with 10% word substitution. This is corroborated by other sources discussing the technology's robustness and effectiveness.
Fact Check Score
None
Fact Check Weight
0
Web Consensus Score
90
Web Consensus Weight
40
Source Quality Score
80
Source Quality Weight
20
Llm Reasoning Score
85
Llm Reasoning Weight
40
Llm Reasoning Score Raw
None
Weighted Total
92
Evidence Summary
Asia Research News confirms BREW's 96.5% detection rate with 10% word substitution.
Sources
asiaresearchnews.com
92
True
Technology
BREW successfully reduced the false positive rate to 2%.
The claim is supported by evidence from Asia Research News, which confirms that BREW reduced the false positive rate to 2%. This is corroborated by other sources discussing the technology's performance and effectiveness in minimizing false positives.
Fact Check Score
None
Fact Check Weight
0
Web Consensus Score
90
Web Consensus Weight
40
Source Quality Score
80
Source Quality Weight
20
Llm Reasoning Score
85
Llm Reasoning Weight
40
Llm Reasoning Score Raw
None
Weighted Total
92
Evidence Summary
Asia Research News confirms BREW's reduction of false positive rate to 2%.
Sources
asiaresearchnews.com
94
True
Technology
The research findings have been accepted by the International Conference on Machine Learning (ICML) 2026.
The claim is supported by evidence from the ICML 2026 blog and other sources confirming the acceptance of the research findings at the conference. This is a credible source, and the information is consistent across multiple platforms.
Fact Check Score
None
Fact Check Weight
0
Web Consensus Score
95
Web Consensus Weight
40
Source Quality Score
85
Source Quality Weight
20
Llm Reasoning Score
90
Llm Reasoning Weight
40
Llm Reasoning Score Raw
None
Weighted Total
94
Evidence Summary
ICML 2026 blog confirms acceptance of research findings.
