🟪 RUSSIAN TEAM USED WESTERN AI TO BUILD AN AUTONOMOUS KAMIKAZE-DRONE SWARM
A Russia-based team used Anthropic’s Claude Code to develop software for a swarm of autonomous FPV kamikaze drones capable of selecting targets—including people—and issuing detonation commands without a human operator, Anthropic revealed on September 10.
The team used Claude to build shared swarm memory, coordination systems, camera-based terminal guidance, attack and return-to-base behavior, acoustic detection and technology intended to locate opposing drone operators.
A separate computer-vision system was trained using scraped footage from combat operations in Ukraine. Targets were divided into “enemy” and “friendly” categories, with Russian military systems specifically protected from attack. The developers also repeatedly used a fixed location in Donetsk Oblast as a simulated strike point.
This was more than a conversation or theoretical proposal. Anthropic found that the developers had loaded code onto real circuit boards and connected physical hardware to a simulated combat environment. However, the company assessed the technology at an early experimental stage and found no evidence that a functioning swarm had reached the battlefield.
The developers bypassed Anthropic’s regional restrictions through commercial virtual servers. Anthropic banned the associated accounts after discovering the weapons project, but the code Claude had written was already saved directly onto the developers’ computers and could continue to be used without access to Claude.
Anthropic assessed that the actors were a small, specialized Russian freelance team with ties to a regional university and a federal research center associated with the Russian Academy of Sciences. The developers claimed to have Russian government and Defense Ministry funding, but Anthropic could not verify those claims.
This is precisely why access to advanced Western AI cannot be treated like an ordinary consumer service. Frontier models can shorten weapons-development timelines, replace scarce engineering expertise and transform battlefield footage into autonomous targeting systems.
Russia and other hostile military actors should not be permitted to use Western artificial intelligence to build weapons against Ukraine or the nations that created the technology. Geographic restrictions alone are clearly insufficient; stronger identity checks, behavioral monitoring and intelligence sharing are now matters of national security.