Anthropic: Russia-linked hackers used Claude Code to automate cyberattacks on Ukraine and Europe

Why it matters

The case is the first publicly detailed large-scale use of a commercial frontier AI to orchestrate a state-linked cyber operation rather than merely assist a human operator. It shows how AI can compress and industrialize the full intrusion chain — reconnaissance, phishing, malware modification, persistence and data theft — while reducing the need for skilled human involvement. It also exposes concrete risk to Ukrainian and European ministries, intelligence agencies, embassies, think tanks and the defense-industrial base, especially drone developers and their suppliers, and to adjacent ecosystems such as hotel Wi-Fi networks and personal messaging accounts, including WhatsApp correspondence of former senior Ukrainian officials.

What happened

Anthropic's report described how a cyber actor it tracks as GTG-20006 — which researchers link to the Russian state-aligned group Midnight Blizzard — used the Claude Code AI tool to automate cyberattacks primarily against Ukraine and European countries. One of the operators is a Russian-speaking hacker known as JackPoterz who communicated with the chatbot in Russian. Between December 2025 and August 2026 the group hit more than 20 organizations, including ministries, defense and intelligence agencies, embassies, diplomatic missions, think tanks and defense-industrial firms; outliers included a Southeast Asian maritime shipping and tracking body and a North African state technology agency.

The attackers built AI-driven workflows that handled target research, domain registration and hosting setup for phishing emails, message delivery, monitoring of command-and-control channels for successful intrusions, and large-scale data exfiltration. A human operator was involved mainly to fine-tune skills in Claude Code when workflows needed adjustment. The AI could independently modify malware: when antivirus flagged a sample, it analyzed why it was detected, altered the code, recompiled the program and repeated the cycle until defenses stopped recognizing it.

Key themes included Ukraine and the developers and supply chains of military drone technologies. Hackers stole data on drone tech, including email inboxes of at least two drone-component makers, and spent several days analyzing a stolen drone machine-vision software suite to reconstruct its architecture, components and suppliers. To deliver malware they compromised at least three hotel Wi-Fi providers and altered DNS records to redirect guests' devices to attacker servers, serving payloads for Windows, Android and iOS. The group also gained access to victims' WhatsApp accounts and mass-downloaded correspondence in Ukrainian and Russian, including accounts of at least two former senior Ukrainian officials.

Background

Midnight Blizzard, also tracked as APT29 or Cozy Bear, is a Russian state-aligned cyber-espionage group previously tied to intrusions against governments and organizations in multiple countries. While threat actors have long used automation and scripting, Anthropic frames this case as a step-change in which AI orchestrates significant portions of an operation rather than merely assisting a human. Military drone development and its supply chain have been a documented focus of Russian intelligence interest during the war in Ukraine.