The government's project to develop homegrown AI foundation models has hit a snag again after claims emerged that a core feature of Naver's multimodal AI model was borrowed from China's Alibaba model. 

According to sources in the tech industry on Tuesday, claims have surfaced that Naver Cloud's vision language model, HyperCLOVA X SEED 32B Think, used the vision encoder of Alibaba's Qwen two-point-five large language models as a key feature. 

Naver Cloud admitted to using Chinese open source, saying it strategically adopted an already-verified encoder after taking into account compatibility with the global technology ecosystem and the efficient optimization of the whole system.

Naver said it borrowed the feature not because it lacks technology but because using verified encoders is the industry standard recognized internationally. 

It said it opted to use the Chinese feature based on strategic assessment and had earlier disclosed its decision in a technical report.

Previously, AI startup Upstage, which is also taking part in the government project, was accused of using a Chinese model. But the individual who made the claim later retracted their statement and issued an apology.