AI · 49 days ago
US agencies map industrial AI capability theft The NSA, CISA, and FBI escalated their warning on China-based AI firms, naming DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI and saying they pulled billions of tokens from Anthropic Claude, OpenAI GPT, Google Gemini, and Grok since at least late 2024. The agencies now frame the activity as industrialized distillation, not a side effect of model use.
The technique is simple in plain terms: one model’s answers are harvested at scale and fed into another model so the student learns the teacher’s behavior without paying to discover it itself. The advisory says the firms used premium subscriptions, proxy routing, and blocking evasions to keep collecting outputs, and maps the tradecraft to MITRE ATLAS as a structured development pipeline.
For frontier-model providers, the durable exposure sits in the service layer: if valuable behavior is reachable through APIs or shared subscriptions, it can be extracted repeatedly even when the underlying model weights stay untouched. The remaining question is not whether distillation exists, but how much competitive advantage and training spend the collection pipeline can strip away before normal abuse controls notice it.
Timeline Sources 7 sources covering this story
Dark Reading Sep 9
US Government Claims Chinese AI Firms Distilling Frontier Models
US agencies claim Chinese companies covertly extracted billions of tokens from OpenAI, Anthropic, Google Gemini, and SpaceX's Grok.
BleepingComputer Sep 9
US says Chinese firms extracted billions of tokens from frontier AI models
cybersecurity and intelligence agencies say that six Chinese AI companies conducted industrial-scale distillation attacks on American frontier AI models since at least late 2024.
SecurityWeek Sep 9
US Agencies Warn China Is Systematically Extracting Frontier AI Capabilities
Distillation is an ‘attack’ against an AI model designed to capture outputs, understand reasoning processes, and subsequently train a different model.
The Hacker News Sep 9
U.S. Agencies Accuse China AI Firms of Distilling Claude, GPT, Gemini, and Grok
agencies accuse six China-based AI firms of industrial-scale distillation to extract proprietary capabilities from frontier AI models.
Help Net Security Sep 9
Chinese AI firms are siphoning capabilities from American models, CISA warns - Help Net Security
agencies warn China-based AI companies are using AI knowledge distillation to extract capabilities from leading American models.
CyberScoop Sep 8
Feds accuse China of ‘systematic’ distillation of U.S. AI models
A joint advisory from the NSA, CISA, and FBI accuses Chinese AI firms of industrial-scale distillation to illegally extract capabilities from top U.S.
CISA Advisories Sep 8
China-Based Artificial Intelligence Companies Conducting Industrial-Scale Distillation Campaigns Against U.S. AI Companies | CISA
China-Based Artificial Intelligence Companies Conducting Industrial-Scale Distillation Campaigns Against U.S.
Entities DeepSeek Moonshot AI Alibaba MiniMax StepFun Z.AI Part of the PlainSec briefing for 2026-09-10
Editions Related stories AI · 49 days ago
US agencies map industrial AI capability theft The NSA, CISA, and FBI escalated their warning on China-based AI firms, naming DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI and saying they pulled billions of tokens from Anthropic Claude, OpenAI GPT, Google Gemini, and Grok since at least late 2024. The agencies now frame the activity as industrialized distillation, not a side effect of model use.
The technique is simple in plain terms: one model’s answers are harvested at scale and fed into another model so the student learns the teacher’s behavior without paying to discover it itself. The advisory says the firms used premium subscriptions, proxy routing, and blocking evasions to keep collecting outputs, and maps the tradecraft to MITRE ATLAS as a structured development pipeline.
For frontier-model providers, the durable exposure sits in the service layer: if valuable behavior is reachable through APIs or shared subscriptions, it can be extracted repeatedly even when the underlying model weights stay untouched. The remaining question is not whether distillation exists, but how much competitive advantage and training spend the collection pipeline can strip away before normal abuse controls notice it.
Timeline Sources 7 sources covering this story
Dark Reading Sep 9
US Government Claims Chinese AI Firms Distilling Frontier Models
US agencies claim Chinese companies covertly extracted billions of tokens from OpenAI, Anthropic, Google Gemini, and SpaceX's Grok.
BleepingComputer Sep 9
US says Chinese firms extracted billions of tokens from frontier AI models
cybersecurity and intelligence agencies say that six Chinese AI companies conducted industrial-scale distillation attacks on American frontier AI models since at least late 2024.
SecurityWeek Sep 9
US Agencies Warn China Is Systematically Extracting Frontier AI Capabilities
Distillation is an ‘attack’ against an AI model designed to capture outputs, understand reasoning processes, and subsequently train a different model.
The Hacker News Sep 9
U.S. Agencies Accuse China AI Firms of Distilling Claude, GPT, Gemini, and Grok
agencies accuse six China-based AI firms of industrial-scale distillation to extract proprietary capabilities from frontier AI models.
Help Net Security Sep 9
Chinese AI firms are siphoning capabilities from American models, CISA warns - Help Net Security
agencies warn China-based AI companies are using AI knowledge distillation to extract capabilities from leading American models.
CyberScoop Sep 8
Feds accuse China of ‘systematic’ distillation of U.S. AI models
A joint advisory from the NSA, CISA, and FBI accuses Chinese AI firms of industrial-scale distillation to illegally extract capabilities from top U.S.
CISA Advisories Sep 8
China-Based Artificial Intelligence Companies Conducting Industrial-Scale Distillation Campaigns Against U.S. AI Companies | CISA
China-Based Artificial Intelligence Companies Conducting Industrial-Scale Distillation Campaigns Against U.S.
Entities DeepSeek Moonshot AI Alibaba MiniMax StepFun Z.AI Part of the PlainSec briefing for 2026-09-10
Editions Related stories