In a joint release, the NSA, CISA, and FBI accused six Chinese AI firms—DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI—of conducting industrial-scale distillation attacks against US frontier models, including Claude, GPT, Gemini, and Grok, beginning as early as late 2024. These activities allegedly allow Chinese companies to bypass significant development costs and accelerate training timelines by extracting proprietary functionalities. The agencies detailed two primary technical methodologies: the exploitation of inference APIs via swarms of fraudulent accounts and the use of advanced prompt injection. In the latter method, actors employ jailbreaking techniques to force models to reveal restricted internal processes, such as step-by-step chain-of-thought reasoning. For instance, DeepSeek reportedly utilized specific prompts designed to compel models to articulate their hidden internal logic. To combat these coordinated campaigns, which utilize a gray market of proxies to circumvent geographical restrictions and route requests through multiple pathways, US agencies are calling for enhanced detection capabilities. The agencies emphasized that mitigating these threats will require coordinated ecosystem-wide action to prevent the unauthorized extraction of restricted AI capabilities that threaten the United States' technological leadership in the field.