US agencies accuse Chinese AI firms of systematically copying American models

Sep 8, 2026

U.S. cybersecurity and intelligence agencies have accused six China-based AI firms, including DeepSeek and Alibaba, of running industrial-scale distillation campaigns since 2024 to systematically copy proprietary features from leading American models like Claude and GPT. Officials allege these campaigns, likely conducted with Chinese government awareness, form the core of the firms' development strategies. Beijing has dismissed the accusations, attributing its AI progress to technological self-reliance.

Chinese AI distillation campaigns

▪The joint advisory alleged that the Chinese companies targeted U.S. frontier models including Anthropic's Claude, OpenAI's GPT, Google's Gemini, and xAI's Grok, with activity dating back to at least late 2024.
▪The joint advisory stated that the distillation campaigns were conducted "likely with Chinese government awareness," though it did not claim that Chinese intelligence played a direct role.
▪The FBI, the National Security Agency, and the Cybersecurity and Infrastructure Security Agency issued a joint advisory on September 8, 2026, accusing six China-based AI companies of running industrial-scale campaigns to systematically extract proprietary features from leading U.S. models.
▪The joint advisory named DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI as the six Chinese firms conducting the systematic distillation campaigns.

Knowledge distillation infrastructure methods

▪The joint advisory stated that Chinese AI firms used a gray market of proxies known as "transfer stations" to bypass geographic restrictions, breach terms of use, and evade safeguards.
▪The joint advisory alleged that Chinese AI firms routed distillation requests through native APIs, remote cloud providers, and third-party aggregators that automatically obfuscate user metadata to avoid detection.
▪In February 2026, Anthropic accused DeepSeek, Moonshot AI, and MiniMax of creating approximately 24,000 fraudulent accounts and sending over 16 million prompts to its Claude model.

DeepSeek training cost claims

▪The joint advisory accused DeepSeek of running an organized campaign since late 2024 to feed synthetic training data into its R1 and V3 models, calling DeepSeek's widely cited $5.6 million training figure misleading because it excludes the cost of distilled data.
▪In early 2026, OpenAI stated that DeepSeek had inappropriately based its model on ChatGPT by using OpenAI's data.

US agency security recommendations

▪The joint advisory recommended that U.S. model providers monitor for anomalous prompts, usage spikes, and immediate maximum usage from new accounts to detect distillation campaigns.
▪The joint advisory recommended that U.S. companies subtly alter or degrade responses to suspected malicious queries to reduce the payoff for companies conducting industrial-scale distillation.

Chinese government response

▪The Chinese government has previously responded to U.S. intellectual property allegations by accusing U.S. companies of distilling Chinese models and issuing veiled threats of retaliation against potential U.S. bans.
▪Chinese Ministry of Foreign Affairs spokesperson Mao Ning stated on September 9, 2026, that China's AI development is a result of high-level technological self-reliance and urged the U.S. to strengthen cooperation rather than make groundless accusations.