China is trying to steal US AI models' secrets, intel agencies warn

The FBI, NSA and CISA warned that six Chinese companies have conducted large-scale efforts to distill capabilities from leading US artificial-intelligence models since late 2024. The agencies named firms including DeepSeek, Moonshot AI and Alibaba and said targets included systems from OpenAI, Anthropic, Google and SpaceX. Model distillation uses outputs from a stronger system to train another model, potentially bypassing safeguards or contractual restrictions when performed at scale. China rejected the allegations, while the advisory urged tighter monitoring, access controls and abuse detection by providers.