China's internet regulator has opened a formal investigation into AI developers DeepSeek and Moonshot AI over allegations that both companies routed sensitive user requests to Anthropic's Claude models without customer knowledge. The probe follows a September 2026 Anthropic threat intelligence report naming seven China-based firms — DeepSeek, Moonshot, Alibaba, Zhipu (Z.ai), SenseTime, MiniMax, and Xiaomi — for mass unauthorized access to Claude.
According to Anthropic, DeepSeek relayed selected user requests to Claude while users believed they were interacting with DeepSeek's own systems, and extracted Claude's hidden reasoning steps, described as its "thinking signature," to help train its own models. During a 14-day period in July, DeepSeek allegedly rerouted more than 12.1 million user queries to Claude. Moonshot AI is accused of forwarding nearly 300,000 customer requests to Claude over 10 days through 5,380 fraudulent accounts, mostly located in Singapore and Japan. Anthropic said it could not confirm whether Moonshot disclosed the third-party routing to customers. Among the forwarded material was information linked to a Chinese technology firm's AI program, records from a Chinese municipal police system, and data associated with a Russian government agency. The Cyberspace Administration of China has questioned representatives from all seven companies and is assessing whether the conduct violated China’s cross-border data rules. Beijing has denied the distillation narrative, framing Washington’s complaints as an attempt to hold back China's AI sector.
Separately, the UN Security Council is set to hear directly from AI builders on Wednesday, including OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, AI safety researcher Yoshua Bengio, and Hugging Face CEO Clément Delangue. DeepSeek and Moonshot AI have also been invited to address the 15-member council. France organized the session; the concept note is said to center on AI systems slipping beyond human control, including risks of AI-driven cyberattacks, election interference, and autonomous weapons. Amodei, appearing remotely, recently argued in an essay titled "We Must Pace the Frontier" that frontier AI development should be deliberately slowed and that the U.S. should maintain advanced chip export restrictions on China. The briefing comes two days before President Trump's scheduled September 24 summit with President Xi Jinping, where AI safety is expected to be discussed alongside trade. For cryptocurrency markets, the developments are largely neutral, with no direct crypto assets or blockchain firms implicated.
