Anthropic Report: Artificial Intelligence Models Used in Weapons and Cyber Attacks
A newly published report revealed that Claude models were misused in conventional weapons development, cyber operations, and various disinformation activities.
The latest report published by Anthropic revealed that Claude artificial intelligence models were misused between December 2025 and August 2026 in areas such as conventional weapons development, cyber espionage, and influence operations.
Conventional Weapons Development Activities
It was determined that Claude models were used in developing software for traditional weapons such as firearms, missiles, armed unmanned aerial vehicles (UAVs), bombs, and targeting control systems.
It was reported that a group in northern Yemen used Claude Code in its missile program, though there was no evidence that an operational weapon was produced, and that a Chinese actor utilized artificial intelligence for an anti-torpedo system.
Cyber Espionage and Attacks
Within the scope of cyber attacks, it was determined that espionage activities belonging to Russia targeted Ukrainian and European military intelligence.
It was noted that China-based actors carried out activities involving infiltration into production systems and reconnaissance targeting foreign government networks, while in France, an individual targeted political parties and media organizations.
Surveillance and Fraud Cases
In surveillance operations, activities targeting Iran and the Persian Gulf by an Israel-Singapore-based entity and China-centric public opinion monitoring tools were deciphered.
In fraud cases, it was identified that a Chinese studio used artificial intelligence characters in dating applications.
Distillation Activities and Disinformation
The report alleged that companies such as Alibaba, Moonshot AI, DeepSeek, and Xiaomi used unauthorized distillation activities and covert methods such as redirecting user requests to Claude.
In the influence operations section, it was reported that Russia-linked activities and France-based disinformation networks generated fake content with the help of Claude.