Cisco Talos published an analysis on October 31 of a Windows infostealer it calls VOLTA-AUTODOXER, which turns a credential theft into a prioritised dossier on the victim. The central caveat comes first: Talos has NOT tracked whether VOLTA has been used in the wild, the AI stage is optional, and the analysed sample looks like a prototype or builder product sold to other attackers. This is an analysis of a design, not a confirmed wave of attacks.
What VOLTA steals
The first sweep takes browser passwords, cookies and saved cards, Microsoft OAuth tokens, crypto wallet seed phrases, Discord, Telegram, Steam and Roblox sessions, cloud-provider keys and saved Wi-Fi passwords. A second sweep uses stolen Gmail access to search receipts — from Amazon, DoorDash, Uber Eats, Grubhub, Walmart and PayPal — and order histories to infer a home address, employer, inferred birth year and the last four digits of cards, and it prices Roblox items via Rolimons.
The AI ranking stage
The bundle is sent through OpenRouter with a prompt asking the model to act as a forensic intelligence analyst drafting a target assessment, with a WHAT THEY’RE WORTH section ranking valuables by monetisation value and a THEIR OPSEC section on security posture. It defaults to Anthropic’s Claude Opus 4.6 and, if the reply contains refusal phrases such as “I can’t” or “I’m not able”, falls back to Moonshot AI’s Kimi K2 and then Google’s Gemini 2.5 Flash. The returned assessment is saved as AI_Analysis.txt. Talos does NOT claim every named model complied — the code contains this failover chain, and that is all that can be said about it.
Why researchers are concerned
The AI step converts a raw credential log into a ranked, citation-backed dossier, and the six swappable build-time slots for Telegram channel, API keys and settings point to a builder model where a buyer of any skill level could receive ranked assessments. That combination — automated theft plus automated triage — is what makes the design worrying even before any real-world use is confirmed, and it fits a wider pattern of AI-era supply-chain and coding-tool threats.
What users can do
Talos’s mitigation note is that blocking unknown processes from launching Chrome with –headless cuts off much of the collection route. As general advice, not a Talos quote: use unique passwords, turn on two-factor authentication, and never store crypto seed phrases in plaintext.
What is still unknown
Whether it has been deployed, which models answered, and how widely any builder version has spread all remain unknown. Until Talos or another researcher documents in-the-wild use, VOLTA should be described as an analysed design, not as an active campaign.
FAQ
Has VOLTA attacked people? Not tracked in the wild — no confirmed attacks have been documented.
Does a model refusal stop it? It stops the AI assessment only; the underlying theft can still run.
Is the AI stage always used? No — Talos describes it as optional, and the sample looks like a prototype or builder product.
Sources
- Unite.AI: Cisco Talos Details VOLTA Infostealer’s LLM Victim Ranking Stage
- Startup Fortune: Malware Called VOLTA-AUTODOXER Uses AI to Build a Dossier on Its Victims
- Superpower Daily: Cisco Talos Finds Malware That Uses AI to Rank Victims’ Stolen Assets
