AI Is Going Just Great
← Timeline
·2w agoScaryMajoranthropic

Windows Malware Uses Claude to Rank Stolen Data by Victim Asset Value Before Self-Destructing

Published · updated · curated by AI Is Going Just Great

Source: tomshardware.com ↗ · archived copy ↗

payload builds custom target assessment, can switch models if one refuses, or skip the AI step entirely

A newly discovered Windows malware strain feeds stolen victim data directly to Anthropic's Claude to generate a prioritized assessment of which assets are worth targeting. The payload identifies files, credentials, and system information, ships them to Claude for ranking, then deletes itself to minimize forensic traces. If Claude refuses the request, the malware is designed to fall back to a different model or skip the AI step entirely and proceed with a generic assessment.

The design treats AI refusals as a minor inconvenience rather than a hard stop. Model-switching fallback logic means no single provider's safety guardrails are a reliable choke point — the malware treats them as optional dependencies.