Anthropic reports state actors used Claude AI for weapons development and espionage

Anthropic has released a comprehensive threat report revealing that state-aligned actors and militant groups have used its Claude AI models for weapons development, espionage, and influence operations. In northern Yemen, a threat cell used Claude to write guidance software for ballistic missiles and rockets, even conducting a failed test-fire. Meanwhile, a campaign linked to the UAE government utilized the AI to generate 300 fake influencer accounts, profile European politicians, and ghostwrite UN testimonies. Anthropic has banned the associated accounts and shared intelligence with partners.

Claude AI weapons development

• ▪A threat actor cell based in northern Yemen used Anthropic's Claude Code to develop guidance, navigation, and flight-control software for three distinct weapons programs
• ▪Anthropic released a threat report detailing how state-aligned and malicious actors used its Claude AI models for weapons development, global espionage, and mass surveillance

Yemen Houthi missile guidance

• ▪The weapons programs in northern Yemen included a guided rocket using a commodity phone-class flight computer, a multi-stage ballistic missile with a range goal above 2,000 kilometers, and a multi-variant missile featuring a hypersonic glide vehicle variant
• ▪The northern Yemen operators test-fired a guided rocket developed using Claude, which failed, prompting the users to return to the chatbot within hours to diagnose the failure
• ▪Hazam al-Assad, a member of the Houthis' political bureau, denied relying on open sources like Claude, stating that Houthi armed forces possess their own developed technology and production capabilities

UAE Muslim Brotherhood influence

• ▪Anthropic detected and disrupted a United Arab Emirates-directed influence operation that used Claude to build a network of approximately 300 inauthentic influencer social media accounts
• ▪A United Arab Emirates-directed actor used Claude to draft official testimonies for the 62nd session of the UN Human Rights Council and create a front NGO mimicking a real Swiss organization
• ▪The United Arab Emirates-directed campaign used Claude to compile detailed personal files on 18 European parliament members and journalists, alongside counter-accountability dossiers on UN Special Rapporteurs

Anthropic account detection bans

• ▪Anthropic banned accounts linked to a Russian espionage operation, suspected to be Midnight Blizzard, which used automated AI workflows to target Ukrainian, European, and diplomatic entities
• ▪Anthropic disrupted a Chinese offensive cyber operation run by university students in Hunan province that used Claude to target government and corporate networks across the Middle East, Europe, and Southeast Asia
• ▪Anthropic banned the accounts associated with the northern Yemen weapons programs and shared threat intelligence with public- and private-sector partners

AI-assisted disinformation campaigns

• ▪A China-aligned account with no Arabic language skills used Claude to run a multi-day recruitment operation to infiltrate Uyghur targets in Syria by drafting outreach and translating replies in real time
• ▪Anthropic identified and removed three Iranian state-aligned accounts using Claude to run covert influence and psychological operations aligned with Islamic Revolutionary Guards Corps narratives

Debatable claims

• ▪AI developers should restrict advanced coding tools to vetted users
• ▪The US Pentagon should blacklist AI developers that restrict military use of their technology
• ▪AI developers should be legally liable when their models are used to design weapons