Anthropic Reports Misuse of Claude for Surveillance, Weapons, and Disinformation
Anthropic published a report documenting misuse of Claude AI for state-sponsored surveillance, weapons development, and disinformation campaigns across multiple countries. The incidents reportedly included a UAE-linked operation targeting UN experts investigating Sudan, a Mali military surveillance system, Russian autonomous weapons development, and Yemeni missile fabrication attempts.
What happened
Anthropic reported discovering multiple coordinated campaigns in which Claude was misused for state-sponsored surveillance, weapons development, and disinformation. A network linked to the United Arab Emirates used Claude agents to generate documents, social media content, and false testimonies about Sudan's conflict, targeting UN experts critical of UAE involvement and impersonating a Sudanese NGO to pass fabricated witness statements to the UN. The campaign operated around a virtual persona called 'Deadshot' and used approximately 300 fake social media accounts.
In separate incidents reported by Anthropic, a technology consultant in Bamako working with Mali's military intelligence used Claude Code to create 'Lakana 360', a surveillance system capable of monitoring approximately 25 million SIM cards on the country's mobile network. In Russia, a development team used Claude to create software for autonomous 'kamikaze' drones that could select targets without human decision-making. In northern Yemen, a group used Claude Code to assist in missile fabrication, including designing guidance software, creating long-range missile plans, and developing hypersonic glide vehicle designs.
Reported impact
Anthropic stated it could not determine whether the false testimonies or files about UN experts reached their intended recipients. The surveillance system created in Mali reportedly remained active and could intercept calls, text messages, and voice traffic, and identify individuals by voice even after changing SIM cards, as well as detect encrypted communication and VPN use.
Anthropic indicated no evidence that the Yemen group completed a functional weapon, though they conducted a test launch of a rocket that reportedly failed. The Russia team reportedly received funding from Russian government and military research programs.
Index analysis
The reported incidents span multiple countries and failure modes, demonstrating post-deployment misuse of Claude across surveillance, weapons development, and information manipulation. Anthropic's ability to stop some operations but not others, such as the Mali surveillance system, highlights limitations in controlling deployed AI systems after release.
Governance controls
Anthropic halted the networks it identified and published a report describing the unauthorized operations. The company used geographic blocking and other measures but noted that users employed VPNs and divided work across multiple sessions to evade restrictions.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.