The White House has accused Chinese artificial intelligence startup Moonshot AI of illicitly distilling capabilities from Anthropic's Fable model to develop its own Kimi K3 model. Michael Kratsios, director of the White House Office of Science and Technology Policy, stated that the U.S. government possesses information indicating this process occurred during K3's development. Moonshot AI allegedly created a sophisticated internal platform to conduct large-scale model distillation, switching between access methods to evade detection. Model distillation is a technique where a smaller AI system learns from a larger, more advanced model. While legitimate uses exist, U.S. officials argue that covert, industrial-scale distillation amounts to intellectual property theft.
China's Foreign Ministry spokesperson Lin Jian responded by asserting that China's AI progress stems from "self-reliance" and "strength in science and technology". He stated that China opposes the "politicizing and instrumentalizing" of trade and technology issues, suggesting such actions hinder global AI advancement. The allegations highlight growing concerns in Washington regarding the methods employed by Chinese AI firms, particularly as the U.S. seeks to restrict China's access to advanced AI chips and computing infrastructure.
In parallel, the U.S. Army has had to reinstate limits on its artificial intelligence (AI) tool usage. Personnel rapidly consumed their allocated AI processing "tokens" for the year, forcing a reversal of a policy that had offered virtually unlimited access. Internal communications revealed that the Army's central token pool for its Ask Sage platform, which provides access to models like ChatGPT and Gemini, was exhausted by mid-June, despite an announcement in May offering unlimited tokens. This rapid depletion has led to the re-establishment of usage caps across departments.
The Ask Sage platform supports administrative tasks, including processing procurement acquisitions and rewriting job descriptions. An Army insider told WIRED that the entire year's supply of tokens for the service was consumed quickly. While an Army spokesperson attributed the drawdown to broad adoption and stated that leadership was actively managing the token pools, the situation underscores concerns about the "overconsumption" of generative AI tools by both corporate and governmental entities. The total number of tokens available and the exact duration of their consumption are subject to some ambiguity, with different contract records suggesting varying allocations. However, the immediate impact has been the reintroduction of usage controls rather than a complete loss of service.
These developments occur amidst broader discussions in Washington about regulating military AI applications. Senator Kirsten Gillibrand has proposed legislation, "The Secure and Accountable Military AI Act," aimed at restricting the Pentagon's use of AI in operations and regulating its application in autonomous weapons, domestic surveillance, and nuclear weapons systems. This legislative effort seeks to establish clear guidelines for AI use in warfare, emphasizing human control and safety.