Chinese artificial intelligence lab Moonshot AI released its Kimi K3 model, triggering an intense international debate over AI distillation techniques and intellectual property rights. According to public statements and industry disclosures, the release prompted swift reactions from U.S. lawmakers, major technology companies, and artificial intelligence developers regarding how smaller models are trained using outputs from advanced frontier systems.
The Debate Over AI Distillation and National Security
Model distillation is a process where developers use answers or work products from an advanced, large-scale artificial intelligence model to train a separate, smaller system. According to Google AI lead Jeff Dean, who spoke on a podcast in February, distillation is a key technique for making smaller models more capable by leveraging the performance of a frontier model. However, the practice has sparked national security concerns in Washington, D.C., and Silicon Valley regarding foreign labs catching up to U.S. capabilities.
White House advisor Michael Kratsios posted on X that Moonshot AI distilled Anthropic’s Fable model to develop the K3 system. According to Kratsios’s post, the Chinese lab developed an internal platform to conduct large-scale distillation against U.S. models while switching between multiple access methods to avoid detection. Pukar Hamal, founder of AI security firm SecurityPal, described distillation by comparing it to copying another student’s homework after someone else did the hard work of reading the textbook and attending lectures.
Tech Giants Push Back Against Premature Restrictions
In response to the growing scrutiny, a coalition of major technology companies, including Nvidia, Microsoft, Meta, and Palantir, joined more than 20 other firms to sign a letter urging policymakers to avoid premature restrictions on open-weight AI models. According to the letter released by the coalition, distillation is a widely used and legitimate technique for model improvement, evolution, and validation that helps drive innovation and lower costs. Box CEO Aaron Levie, a signatory of the letter, stated in an interview that U.S. companies need access to the best global technology to stay competitive and keep progress moving forward.
Industry research supports the widespread adoption of the practice. Shashi Bellamkonda, research director at Info-Tech Research Group, noted that distillation is practiced regularly across the industry, pointing out that Nvidia used the technique during the training process for its Llama Nemotron series of models.
Anthropic Raises Industrial-Scale Concerns
Despite broad industry use, leading model developers take a stricter stance to protect their commercial investments and platform safety. In February, Anthropic stated that its Claude capabilities were being distilled on an industrial scale by Chinese labs including DeepSeek, Moonshot, and MiniMax, utilizing approximately 24,000 fake accounts to generate 16 million exchanges. Anthropic argued that stopping unauthorized distillation is a matter of national security because U.S. companies build systems designed to prevent malicious actors from utilizing AI to develop bioweapons or conduct cyberattacks. Consequently, companies like Anthropic and OpenAI prohibit distillation in their terms of service, treating unauthorized use of their larger models as potential intellectual property theft.
At the same time, copyright litigation complicates the intellectual property debate. Max Pritt, an attorney for Boies Schiller Flexner representing book authors in copyright lawsuits against AI firms, noted that major technology companies have built their own models using unauthorized sources of content, leaving the government in a complex position as it attempts to protect corporate intellectual property while facing ongoing copyright claims from creators.
Worth a look