The U.S. government is reportedly investigating the Chinese startup Moonshot AI, which recently unveiled a high-performance artificial intelligence (AI) model, to determine whether it has used advanced AI semiconductors to evade U.S. regulations. Allegations have also emerged that the company may have unlawfully utilized models from American AI firms, raising the possibility of sanctions.
According to a report by the technology news outlet The Information on July 22, the U.S. Department of Commerce's Bureau of Industry and Security (BIS) is examining whether Chinese AI companies, including Moonshot AI, have employed advanced AI semiconductors that the U.S. has restricted from being sold to China.
U.S. authorities are investigating whether these companies have circumvented export controls through overseas subsidiaries or third-party data centers. If violations are confirmed, the companies could be added to the BIS's Entity List, which would restrict their access to U.S. products and technologies.
Moonshot AI is under scrutiny for allegedly using NVIDIA's latest AI semiconductors during the development of its recently released AI model, 'Kimi K3.'
Michael Kratsios, the White House Chief Technology Officer, stated on social media that Moonshot AI has secured servers equipped with NVIDIA's GB300 and has reportedly used the GB300 in Thailand. This raises concerns about its potential use in training AI models.
The GB300 is NVIDIA's latest AI system based on the Blackwell architecture, which is restricted from being sold to Chinese companies.
There are also allegations that Moonshot AI has unlawfully utilized models from American AI firms. Kratsios claimed that Moonshot AI has built an internal system capable of extracting large amounts of responses from the Anthropic AI model to train its own model.
This method, known as 'distillation,' involves using responses from high-performing AI models as training data to enhance the performance of other models. While distillation is a common AI development technique, using data from another company's model without permission could lead to violations of terms of service or intellectual property infringement.
Anthropic has previously claimed that Moonshot AI, along with DeepSeek and MiniMax, has unlawfully collected responses from its AI model 'Claude' for model development.
U.S. Treasury Secretary Scott Vilsack warned that if it is confirmed that foreign AI companies have infringed on the intellectual property rights of U.S. firms, sanctions could be imposed.
However, allegations of export control violations or unauthorized distillation against Moonshot AI have not yet been confirmed, and the U.S. Department of Commerce has not made any official announcements regarding the investigation or potential sanctions.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.