VOLTA, discovered in October 2026, shows criminal infrastructure operationalizing AI. The infostealer harvests credentials, payment data, and cryptocurrency wallets from Windows machines, then sends the collected data to Claude Opus 4.6 via OpenRouter (with fallbacks to Kimi and Gemini) asking it to rank victims by monetization value.
The sophistication is architectural. VOLTA builds deliberate failover logic, suggesting developers expected LLM refusals and designed around them. The failover chain itself, Claude to Kimi to Gemini, reveals developers understand safety measures exist and anticipated encountering them.
The choice of OpenRouter’s API exposes their threat model: no attacker-owned infrastructure, just commodity cloud APIs and malware-as-a-service distribution. The LLM becomes the triage operator, deciding which victims merit further pursuit. If Claude declines, the request pivots to Kimi; if that fails, Gemini. The cloud model is now essential to the attack chain.
This escalates beyond LLMs as tools or social engineering vectors. They’re part of the core attack logic, and threat actors are designing around the guardrails they expect to encounter.
Sources: Appraisal by AI: Model-Driven Victim Prioritization in the VOLTA Infostealer
Coverage: New VOLTA Infostealer Uses Claude, Gemini, and Kimi AI to Rank Victims by Financial Value
Related on this blog: [AUTO] Claude Chains Exploits Across Browsers and Operating Systems • [AUTO] The Exploit Window Collapsed • [AUTO] When Open Weights Enable Exploit Generation