VOLTA Infostealer Turns Stolen Data Into AI-Ranked Victim Dossiers

A new Windows infostealer is using artificial intelligence for more than stealing passwords. VOLTA-AUTODOXER, also called VOLTA Infostealer, collects a broad set of personal and business data, then sends the results through a large language model that ranks the victim’s most valuable assets.

Cisco Talos researcher Ryan Fetterman published an analysis of the malware on October 31, 2026. Talos describes VOLTA as a mature stealer implant with an optional AI layer, while Fetterman says its main theft functions still operate without that extra step.

VOLTA Builds a Detailed Picture of Its Victims

VOLTA starts with a broad credential-theft pass. It extracts browser credentials, cookies, and saved payment cards from Chromium-based browsers, Firefox, and Thunderbird. The malware also collects Outlook profiles, Slack workspaces, Steam and Battle.net sessions, Discord tokens, Telegram sessions, Roblox accounts, Microsoft OAuth tokens, SSH files, cloud-provider configuration files, Wi-Fi profiles, and screenshots.

Its reach extends to gcloud, OpenVPN, and Hide.me configuration files. Rather than simply copying cryptocurrency wallets, VOLTA runs an embedded cracker against each wallet vault during the same pass and records the result on a WALLET VAULTS CRACKED line in its report.

The malware also makes a direct WlanGetProfileList call to dump every saved Wi-Fi password on the host in plaintext. That detail matters because the stolen information is not treated as a pile of unrelated files. VOLTA uses it as the starting point for a second round of investigation.

During that live enrichment pass, the malware reuses stolen sessions. With authenticated Gmail tokens, it searches for receipts from Amazon, DoorDash, Uber Eats, Grubhub, Walmart, and PayPal to help confirm a home address. It pulls Uber Eats and DoorDash order history, enriches a Discord profile, and checks the value of a Roblox inventory through a live lookup on Rolimons.

From Credential Log to Intelligence Report

VOLTA assembles the collected material locally into a structured evidence bundle. The bundle can include a location and home address, employer and employer domain, inferred birth year, card last-four digits, Slack workspaces, document authorship metadata, and iPhone backup indicators.

The evidence is saved as Identity_Report.txt and serialized as JSON. The model’s response is saved separately as AI_Analysis.txt. Exfiltration can happen in three staged archives: volta_log.zip after the stealer runs, volta_dox.zip after the AI call, and volta_doxnet.zip if the network enrichment produces anything.

Fetterman described the unusual part of the process this way: “It compiles all of those things together, ultimately producing a serialized dossier that it can send to a couple LLM models. It’s asking the model to act like an intelligence analyst and produce structured citation-backed assessment of the victim. It’s like a prioritization and analysis function.”

That makes the AI stage different from a malware feature that simply automates an existing task. In Talos’s assessment, the model works as an intelligence analyst for the operator, turning raw stolen data into a prioritized assessment of what the victim may be worth.

The AI Layer Uses Multiple Models

VOLTA sends its prompts through OpenRouter, which can connect the malware to multiple AI providers. Its default request goes to Anthropic’s Claude Opus 4.6, with Moonshot AI Kimi K2 as a fallback and Google’s Gemini 2.5 Flash as another fallback if Kimi K2 fails.

The creators included a 240-word prompt for the analysis. It begins, “You are a forensic intelligence analyst drafting a target assessment from a structured evidence bundle,” then asks for an “executive summary” covering what the person does, “what they’re worth,” and a summary of “their opsec.”

The prompt also tells the model to “connect dots across fields,” including checking whether a Wi-Fi password matches the victim’s last name. “Go deep, not wide,” it says.

Fetterman says the prompt is “pretty carefully crafted not to imply anything overtly malicious.” He also says the creators expected model refusals and built a cascading failover system that moves toward more permissible models. The implant can still run without the AI layer, which Fetterman called “kind of above and beyond,” rather than essential to the theft operation.

CAIRN Finds Malware Through Its AI Clues

Talos found VOLTA through CAIRN, the Cognitive Artifact Intelligence Research Network. Talos released and open-sourced the toolkit on September 22, 2026, describing it as a way to hunt, classify, and track AI-integrated malware through metadata artifacts such as embedded prompts, provider endpoints, and API keys.

CAIRN works without downloading or executing the underlying binaries. Talos said its first findings would arrive over time, beginning with CLOSEDQUORUM, a fully autonomous multi-model consensus orchestrator with no human operator.

Talos says every endpoint observed in the VOLTA work was controlled by a third party, with no attacker-owned infrastructure. Fetterman also said Talos did not track whether VOLTA has been used in the wild.

Still, the design points to a change in how infostealers can be used. Passwords and tokens can expire, and passwords can be changed, but an AI-generated profile can help an operator decide which stolen victim deserves attention first. As Fetterman put it, “We’re seeing more advanced capabilities accessible to lower-level sophistication attackers.”

Stay connected via Google News

Follow us for the latest AI News