Anthropic Exposes State-Linked AI Misuse: Russian Espionage and Chinese Model Distillation Campaigns Target Claude
Anthropic's official report, corroborated across major outlets, reveals Russian espionage and Chinese distillation attacks on Claude, plus weapons-related misuse, highlighting AI's dual-use risks in state and criminal contexts.
Anthropic's September 2026 Threat Intelligence report details the disruption of multiple malicious campaigns using its Claude AI models between December 2025 and August 2026, including operations tied to Russian threat actors and Chinese AI laboratories. The company identified sophisticated, multi-agent AI frameworks for reconnaissance, exploitation, and data exfiltration in cyber operations, far beyond simple chatbot queries.[1][2]
A Russia-linked group, with tradecraft consistent with Midnight Blizzard (also known as APT29 or Cozy Bear), allegedly used Claude to automate aspects of espionage against over 20 Ukrainian, European, and U.S.-linked diplomatic, defense, and intelligence targets. Tactics included phishing, hotel Wi-Fi hijacking, and WhatsApp takeovers; AI helped refine malware evasion and reverse-engineer stolen drone software.[2][3]
Chinese entities conducted large-scale 'illicit distillation' attacks, with operators linked to Alibaba generating over 151 million exchanges in one campaign (May-July 2026) to boost Qwen models. Moonshot AI reportedly rerouted customer queries—including sensitive data—to Claude via proxy networks of fraudulent accounts, displaying responses as its own Kimi outputs. Similar efforts involved DeepSeek, Xiaomi, and others across seven labs.[1][4]
New misuse categories included conventional weapons development: a Yemen-based cell used Claude for guided rocket and missile guidance software; China-linked actors advanced anti-torpedo systems; Russian freelancers researched autonomous drone swarms. Additional cases involved surveillance, propaganda, and biological research attempts.[5][6]
Anthropic disrupted all identified activity, banned accounts, strengthened safeguards, and shared intelligence with partners. The report underscores evolving AI-enabled threats while noting no involvement of its most advanced models in most cases. This aligns with prior Anthropic disclosures on distillation from June 2026 onward.[7]
[Anthropic Threat Intel]: AI misuse by state actors will accelerate proxy conflicts and tech espionage, forcing frontier labs into deeper collaboration with governments on export controls and real-time detection.
Sources (6)
- [1]Anthropic Threat Intelligence Report(https://www.anthropic.com/threat-intelligence)
- [2]Anthropic disrupts Russian, Chinese AI campaigns targeting Claude - Nikkei Asia(https://asia.nikkei.com/business/technology/artificial-intelligence/anthropic-disrupts-russian-chinese-ai-campaigns-targeting-claude)
- [3]Anthropic caught Russia-linked spies using Claude in hacking operations - The Record(https://therecord.media/anthropic-russia-hackers-claude)
- [4]Anthropic details how Claude was misused for surveillance and weapons - The Next Web(https://thenextweb.com/news/anthropic-claude-misuse-threat-intelligence-report)
- [5]Chinese AI labs secretly used millions of Claude exchanges to train their models - CNBC(https://www.cnbc.com/2026/09/11/chinese-ai-labs-moonshot-deepseek-alibaba-anthropic.html)
- [6]Anthropic Says Russian and Chinese threat actors used Claude as a weapons engineering assistant - Business Insider(https://www.businessinsider.com/anthropic-says-threat-actors-used-claude-for-weapons-2026-9)