Upstage AI Plagiarism Controversy Settled... Science Minister Says "Bright Future for Korean AI"

Allegations of Plagiarism in Independent AI Model Resolved Through Public Verification and Apology
Government: "Innovation Through Rigorous Scrutiny"... First Evaluation Results to Be Announced on the 19th

The controversy surrounding Upstage, one of the elite teams in the independent artificial intelligence (AI) foundation project, over alleged plagiarism of a Chinese model, was settled after three days. As Upstage initiated a public verification process, the party who raised the allegations issued an apology. The government assessed this debate as a necessary step for Korean AI to reach greater heights.

Kim Sunghoon, CEO of Upstage, is conducting a public verification regarding plagiarism allegations of 'Solar Open 100B' on the 2nd at the Korea Science and Technology Center in Gangnam-gu, Seoul, inviting industry insiders. Photo by Upstage

On January 3, Vice Prime Minister and Minister of Science and ICT Baek Kyunghoon stated on Facebook, "At the site of the public verification, a variety of opinions were exchanged regarding the originality and appropriateness of Upstage's AI model development approach." He added, "Upstage presented evidence based on data, and experts engaged in healthy questioning and discussion for the AI ecosystem."

He continued, "Innovation becomes stronger through transparent and rigorous verification," and added, "While watching the intense technical debate that heated up our AI industry over the past two days, I actually saw a bright future for Korean AI." He expressed his gratitude to Sunghoon Kim, CEO of Upstage, and Seokhyun Ko, CEO of PsionicAI.

Ha Jungwoo, Chief of Future AI Planning at the Presidential Office, also commented, "I agree 100% with Vice Prime Minister Baek. The technical and constructive discussions and debates, as well as the process leading to a clean resolution over the past two days, are a clear demonstration of our AI ecosystem's global competitiveness."

The controversy began with a LinkedIn post by CEO Ko on January 1. He claimed that Upstage's 'Solar Open 100B' was derived from 'GLM-4.5-Air' by Zhipu AI, a Chinese company. By comparing the 'LayerNorm' parameters, which are part of the model's neural network architecture, he pointed out that there was a high similarity of 96.8% in certain sections. In such a case, it would violate the 'from scratch' criteria required for independent AI foundation projects that must develop models on their own.

On January 2, Upstage immediately responded by holding a public verification session at the Korea Science and Technology Center in Gangnam-gu, Seoul. Refuting CEO Ko's claims, Upstage disclosed the entire training record of the model to prove its 'from scratch' methodology. The verification was broadcast live on YouTube.

CEO Kim stated, "Most large language models (LLMs) are standardized based on either the Transformer or Mixture of Experts (MoE) architectures," and explained, "The cosine similarity of LayerNorm parameters would inevitably be high not only for Solar but also for any AI model." He compared this to opening two different English dictionaries and pointing out that their contents are similar, arguing that this is not an accurate basis for comparison.

He added, "While healthy debate and the exchange of opinions are welcome, asserting false information as fact seriously undermines the efforts of Upstage and the government, who are striving to become one of the top three in AI." He demanded a public apology from CEO Ko.

In response, CEO Ko issued a statement the following day, January 3, saying, "Since this model is being discussed at a national level, I judged it appropriate to quickly bring the matter to public attention before conducting further internal and cross-verification." He continued, "I sincerely apologize to CEO Kim of Upstage, all related parties, and industry professionals who may have been taken aback by the sudden nature of my allegations."

Meanwhile, the government announced that the first evaluation of the independent AI foundation project, targeting elite teams from Naver Cloud, Upstage, SK Telecom, NC AI, and LG AI Research Institute, will be conducted from January 5 to 15, with the results to be announced on the 19th. One team will be eliminated in this first evaluation.