Malware uses LLMs to rank victims by how much they're worth
Points and comments are a snapshot, not live.
New malware uses LLMs to profile and rank victims by monetization value.
VOLTA-AUTODOXER is a Windows infostealer that adds an LLM-powered victim-profiling stage after credential theft. It harvests browser passwords, cryptocurrency wallets, sessions from multiple platforms, and Wi-Fi passwords, then enriches the data via stolen sessions (Gmail receipts, Uber Eats orders). The structured evidence bundle is sent to OpenRouter with a system prompt asking the model to act as a forensic analyst, producing a target assessment with a "WHAT THEY'RE WORTH" section ranked low/medium/high and an "OPSEC" assessment. It uses a multi-model architecture (Claude default, Kimi and Gemini fallbacks) with temperature 0.2 for deterministic output, anticipating refusals.
The malware is self-deleting with no attacker-owned infrastructure, suggesting a builder model where customers get custom-compiled binaries. The LLM integration displaces the operator's analytical work, not improving speed but reducing human attention required.
What commenters are saying
Commenters are primarily concerned about the growing attack surface from AI-assisted malware and rapid software updates. Many debate practical countermeasures: using separate machines for different activities (banking, browsing, gaming), VMs, Docker containers, and QubesOS. Several recount firsthand experiences with coding agents like Claude scanning home directories for code and configurations, illustrating that AI tools themselves already pose data exposure risks. A worker expressing skepticism about LLM quality for prioritization is countered by pragmatists noting it is "good enough" and can be iteratively improved. There is broad unease about supply chain risks in the AI era.