Anthropic says Claude was used in Russian cyberattacks
Anthropic said that groups linked to the Russian Federation — an aggressor state — used the Claude artificial intelligence model for cyberespionage against Ukrainian and European targets, disinformation campaigns in Africa and Moldova, and the development of software for FPV drones. Euronews reports, citing a report by the U.S. company.
Anthropic said it had blocked all identified malicious activity, strengthened its security systems, and shared information with government authorities and industry partners.
Cyberattacks against Ukrainian targets
According to Anthropic, the hacking group GTG-20006 automated operations against more than 20 organizations. The targets included government ministries, defense and intelligence agencies, embassies, think tanks, and defense industry enterprises.
The primary target was Ukraine, including suppliers of military drone technology and their supply chains. The group also attacked government institutions in Middle Eastern and Asian countries connected to maritime activities. Anthropic links GTG-20006 to Midnight Blizzard, a group that Western intelligence agencies attribute to Russia’s Foreign Intelligence Service.
Claude was used to create malware, phishing kits, and surveillance tools. The model was also used to execute commands in compromised networks, collect credentials, and exfiltrate data under human control and, in some cases, with minimal human oversight.
More current news is available on the UA.News Telegram channel Telegram.
Disinformation and drone development
Together with researchers from the All Eyes On Wagner project, the company identified and removed an account of a Russian-speaking operator in Bangui who was preparing pro-Russian content for the Radio Lengo Songo radio station in the Central African Republic. According to investigators, the operator coordinated the preparation of materials with RT, Sputnik Afrique, TASS, and the local Russian House center.
In Moldova, Claude was allegedly used by a former regional head of the Russian state media network Sputnik, according to Anthropic. He adapted news, poll results, and opposition materials to fit pro-Kremlin narratives.
Separately, Anthropic blocked a group of developers linked to Russia who used Claude Code to create software for kamikaze drones. This included an autonomous guidance system that was tested on real hardware. The developers trained models on videos of combat operations in Ukraine and used a fixed location in Ukraine’s Donetsk region as a demonstration target for a strike. The stated goal of the project was a fully autonomous swarm of FPV kamikaze drones.
The company linked nine accounts involved in this project to a regional Russian university and a federal research center affiliated with the Russian Academy of Sciences. According to Anthropic, access to Claude in Russia was circumvented through commercial virtual private servers.
