Anthropic Says Russian Developers Used Claude for Kamikaze Drone Software

Summary

Anthropic says a small team of likely freelance developers in Russia used its Claude chatbot to develop software for an autonomous swarm of first-person-view “kamikaze” drones. The software reportedly included terminal guidance, target selection, coordination among multiple aircraft, and the ability to choose a target and crash into it without a human making the final decision. The developers repeatedly selected a location in Ukraine’s Donetsk region and used VPNs to bypass Anthropic’s geographic restrictions. The findings appear in a 154-page report on misuse of Claude in state-sponsored surveillance and propaganda operations. Anthropic also says cybercriminals and state-backed hackers are using AI to automate much of the attack process. One group allegedly targeted Ukrainian government, military, and diplomatic sectors through phishing, hotel Wi-Fi hijacking, and WhatsApp takeovers. It reportedly used AI to detect when malware was blocked and rewrite the code until it evaded security defenses; Reuters said the group’s methods resembled those of Russia-linked threat actor Midnight Blizzard, which the US has connected to the SVR. Russia’s embassy in Washington did not immediately comment. The report comes amid intensified Russian attacks on Kyiv, including jet-powered drone strikes on three Ukrnafta petrol stations, two of which killed at least two people and injured 12 on Friday, according to Kyiv’s mayor. The briefing also reports casualties in several Ukrainian regions, Ukrainian drone strikes in Russia, upcoming US congressional action on Russian sanctions, and renewed Russian participation in G20 meetings, but these developments are separate from Anthropic’s AI findings.