Most stealers hand an operator a pile of credentials and let the buyer sort it out. VOLTA-AUTODOXER does the sorting in-house.
Cisco Talos documented the Windows malware assembling what its researchers call an evidence bundle about the person behind the machine: home address, employer and its domain, an inferred birth year, card last-4 digits, Slack workspaces, document authorship metadata and a live Roblox account valuation. It then uses stolen session access to search Gmail for Amazon, DoorDash, Uber Eats, Grubhub, Walmart and PayPal receipts, corroborating the address it already guessed.
Three models, one refusal clause
The bundle is serialized and routed through OpenRouter with a 240-word system prompt casting the model as a forensic intelligence analyst. Requests default to Anthropic’s Claude Opus 4.6, falling back to Moonshot’s Kimi K2 and Google’s Gemini 2.5 Flash.
That chain exists for a specific reason. The malware scans each reply for phrases like “I can’t” or “I’m not able” and moves to the next model. The prompt also forbids the model from supplying any fact beyond the bundle, closing the usual hallucination gap.
Output arrives in sections titled “WHAT THEY’RE WORTH” and “THEIR OPSEC,” with valuables tiered low to high and an assessment of the victim’s password hygiene.
What it changes for defenders
Exfiltration runs over Telegram in three stages, the last of which carries the AI dossier. A six-slot build template suggests a builder kit is being sold, meaning the triage capability is not the author’s private edge.
Talos frames the shift plainly: the gain is not speed or scale but offloading analytical work from the operator.
