Anthropic reports foreign actors misuse Claude for espionage, cyberattacks, weapons
Anthropic reported that state and criminal actors from China, Russia, Iran, and Yemen used its Claude AI system to conduct cyberattacks, surveillance, and develop weapons systems. The company found actors used fake accounts and VPNs to bypass access restrictions and halted all cases identified.
What happened
Anthropic reported that criminal hacker groups, Chinese security officials, Russian spies, and Yemeni weapons manufacturers used fake accounts and VPNs to access Claude despite access restrictions being in place.
Over eight months, the actors reportedly used Claude to automate cyberattacks, spread propaganda, conduct mass surveillance, and design advanced military devices. Chinese and Iranian actors used Claude to analyze social media to identify potential dissenters and sensitive political information. A Russia-based actor attempted to use Claude to build autonomous kamikaze drones.
Reported impact
Anthropic halted activity by a group in North Yemen using Claude to develop guided rockets, tiered ballistic missiles, and hypersonic launch missiles. The company discovered five cases where foreign scientists used Claude to research dangerous pathogens, including deadly viruses and highly pathogenic bird flu.
Index analysis
Anthropic Threat Research Head Jacob Klein stated that previously hypothetical misuse cases had now become real, though the cases did not involve Claude Mythos, the company's most powerful AI model.
Governance controls
Anthropic said it halted all identified cases, reported them to government and industry groups, and strengthened Claude's security systems.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.