South Korea’s Dokpamo programme faces scrutiny as decision AI expands

The state audit was scheduled to run from Oct. 6 to Oct. 30, and government AI policy was expected to be a key issue.[1] Likely questions included how Dokpamo and the frontier AI programme divide responsibilities, and whether Dokpamo can be sustained.[1] One speaker asked how the two programmes would be differentiated, while a question also focused on why the 4.7 trillion won frontier AI plan emerged during Dokpamo’s rollout.[2][3] Lawmakers raised concern that the frontier programme could effectively replace Dokpamo.[3] The government said it would continue Dokpamo on schedule and disclose a specific frontier-programme plan later.[1]

Data submitted to a lawmaker showed that 12 of the 31 AI models introduced for the government platform were domestic and 19 were foreign.[4] The government acknowledged that, when the platform was built, it had no criteria for checking the countries or institutions behind the models.[4] It said it does not restrict models by country and selects them based on functionality, performance and security.[4] The list included Bge-M3, developed by Beijing Academy of Artificial Intelligence and added to the U.S. Commerce Department’s Entity List.[4] Bge-M3 converts text into vectors rather than generating answers to user queries.[5]

Dokpamo is a project to develop an in-house foundation model through domestic companies, with the stated aim of strengthening national AI competitiveness.[2] The government said the third Dokpamo evaluation would be completed as planned.[2] Work was also underway to apply four models from the project’s second evaluation to the pan-government common AI base.[3] The government described frontier AI as a way to break South Korea’s technological limits and Dokpamo as one of two pillars for AI use across industry and society.[2] Bae confirmed he was the final person responsible for the government’s national AI programme.[3]