Anthropic has released a 154-page report on the misuse of its artificial intelligence. According to the document, freelance developers from the Russian Federation used the Claude model to create the basic software for a swarm of autonomous kamikaze drones capable of selecting targets — including people — without the involvement of an operator.
Anthropic has released a 154-page report on the misuse of its artificial intelligence. According to the document, freelance developers from the Russian Federation used the Claude model to create the basic software for a swarm of autonomous kamikaze drones capable of selecting targets — including people — without the involvement of an operator.
According to Anthropic, a group called GTG-27005 used the Claude Code tool to write, test, and save code directly into their project files. They bypassed Anthropic's geoblocks using commercial VPS servers. The operation, codenamed DronDoc or Serafim, began in mid-May 2026, it says . Cybernews.
Artificial intelligence helped the enemy implement a shared “swarm memory” and a fail-safe coordination system for group deployment of drones. A small language model was installed on board the UAV, which decided whether the drone should attack, conduct surveillance, or return to base.
The target recognition system was designed for "autonomous lethal destruction." It is capable of independently selecting an object and giving a command to detonate without a human in the decision-making chain. The developers explicitly included the item "human" in the target category.
The Russians have repeatedly used the location in the Donetsk region as a demonstration strike point, and the frontline cities of Ukraine have served as a geological base for missions.
Anthropic believes that a small specialized team of freelancers is behind this, and not directly the Russian government. Specialists discovered 9 accounts associated with a regional university and a federal research center of the Russian Academy of Sciences (RAS). All discovered accounts were blocked.
The developers themselves claimed to have received funding from the Russian Fund for Advanced Research, the National Technology Initiative, and the Russian Ministry of Defense, but Anthropic was unable to verify this data.
In addition, the report notes that Russian-linked hackers (in particular the Midnight Blizzard group) are increasingly using AI for cyberattacks, phishing, and hacking into the accounts of Ukrainian officials and military personnel.
