South Korea's Ministry of Science and ICT will operate a standing monitoring system in which citizens directly identify AI error and risk cases.

The ministry, together with the AI Safety Institute, held a launch ceremony on Wednesday at the Glad Hotel in Yeouido, Seoul, for the citizen-participation monitoring team called AI Safety Watch.

AI Safety Watch is a monitoring team in which citizens find and report errors and risk cases in generative AI. Eighty people were finally selected out of 138 applicants who applied in two open recruitment rounds in August and September. They will carry out AI safety monitoring activities for about a year through August next year.

The main inspection areas are four categories: hallucinations that generate incorrect information or figures as if they were facts; bias and discrimination against certain groups; security vulnerabilities such as bypassing safety devices and exposure of personal information; and technical defects such as no response or repetitive answers.

Cases reported by the monitoring team will be verified based on accuracy, risk level, reproducibility and usability. Verified and confirmed cases will be used as basic data for AI safety evaluations and risk research. The ministry also plans to share the cases with relevant companies.

Earlier, the ministry and the institute ran the "Real AI Challenge with No Information Errors" for about 30 days starting in December last year. Of 625 reports in total, 464 were confirmed as valid error cases. The new monitoring team was set up to expand the temporary campaign into a standing inspection system.

Eom Yeol (엄열), director-general for AI Policy at the ministry, said: "As AI has become part of daily life used by everyone of all ages, a safe environment for use is something we must all create together." He added: "Each error found from the public's perspective will become the foundation for making AI safer."