Malware devs abuse Anthropic’s Claude AI to build ransomware

Anthropic’s Claude Code has been reportedly abused by threat actors to help develop ransomware and support data extortion campaigns, with UK-based operators tracked as GTG-5004 using Claude to create a ransomware-as-a-service operation.

The package reportedly includes a ChaCha20 stream cipher with RSA key management, shadow copy deletion, targeted file options, and the ability to encrypt network shares, while evasion features include reflective DLL injection, syscall invocation techniques, API hooking bypass, string obfuscation, and anti-debugging.

In another case tracked as GTG-2002, Claude was used as an active operator to conduct a data extortion campaign against at least 17 organisations across government, healthcare, financial and emergency services sectors, performing network reconnaissance, initial access, and malware generation based on the Chisel tunneling tool. The ransom demands were set between $75,000 and $500,000, and Claude helped analyse stolen data to set figures and generate custom HTML ransom notes displayed on victim machines.

According to Anthropic, the actor’s dependency on Claude for implementing the most knowledge-demanding bits of the RaaS platform was near-complete, with the operator seemingly unable to implement encryption algorithms, anti-analysis techniques or Windows internals manipulation without Claude’s assistance. The company has banned implicated accounts, shared indicators with partners, and highlighted how AI misuse can enable sophisticated cybercrime.