A new report highlights widespread AI usage in South Korea’s public-sector research, revealing 6,537 AI-related cases from 2023 to the present, with total spending around 1.496 billion won and an average cost per use of about 228,837 won.

The same exclusive findings indicate ongoing AI involvement in national institutes, underscoring a need to scrutinize procurement, usage, and support for Generative AI like ChatGPT.

Some institutes deny illegal AI use, arguing that no reliable safety mechanisms exist to detect violations.

There is a governance gap in AI use for national research bodies, prompting calls for formal safety protocols, oversight, and standardized guidelines across institutes.

Experts urge comprehensive AI safety standards to protect objectivity and confidentiality, advocating a clear scope of use with unified guidelines across national institutes.

Leading universities and state-run institutes criticize AI detection tools as unreliable or biased, with Seoul National University pausing a commonly used detector in mid-year due to accuracy concerns.

Reports note AI-generated errors such as hallucinations, misattributions, and citation mistakes in various outputs, including overseas training results and national budget statistics.

There have been 11 notable AI-related inaccuracies among 1,385 overseas training reports, illustrating the risk of AI-generated misinformation.

Many institutions lack dedicated AI ethics or utilization guidelines; only a few, like the Korea Research Institute of Land, Infrastructure and Transport, reportedly use AI-detection services.

Numerous institutes lack safety measures, with several relying on self-checks or low-accuracy detectors, and some reporting no AI-ethics framework at all.

The data come from Sohn Myung-soo of the Democratic Party, referencing information from government-linked research associations and 26 national institutes.

Breakdowns show AI-provider usage: 4,613 OpenAI (ChatGPT) cases, 1,244 Anthropic (Claude) cases, and 456 Google (Gemini) cases.