Moonshot AI Faces Theft Claims Over Kimi K3 Model

Michael Kratsios, a White House AI adviser, recently alleged that Chinese startup Moonshot AI stole technology from the U.S. company Anthropic. According to these claims, Moonshot used a process called distillation to copy the behavior of Claude, an AI model developed by Anthropic, to build its own Kimi K3 platform. This technique involves training a new model on the outputs of an existing, more powerful one.

Anthropic previously reported that it tracked millions of suspicious requests originating from fake accounts linked to Moonshot staff. These requests reportedly targeted specific features like coding and visual reasoning. The accusations also mention that Moonshot allegedly bypassed U.S. export controls by accessing high end NVIDIA chips through operations based in Thailand.

Moonshot launched Kimi K3 on July 16, quickly becoming a popular open weight model with massive user interest. However, concerns are rising as officials fear that copied models lack necessary safety guardrails. While the U.S. currently leads in overall AI spending, these disputes highlight a growing tension over intellectual property and international trade restrictions.

Washington has not yet announced official penalties regarding these specific claims. The market is now waiting to see how U.S. regulators respond to these ongoing accusations of industrial scale theft. As the technology behind these models becomes more accessible, the battle to protect American AI research remains a top priority for government leaders.