VOLTA Malware Asks AI to Rank Its Victims: Cisco Talos Finds Infostealer That Sorts Stolen Data

Writer Published Updated 0 comments
Illustration of stolen digital files being sorted into ranked folders by an automated system, no text or logos

Why this matters: this malware does not just steal more data — it hands the sorting job to an AI model, so even a low-skill operator could receive a ranked list of who is worth targeting next. What to watch: whether other stealers copy this refusal-resistant, multi-model design.

November 1, 2026 — Cisco Talos has revealed a Windows infostealer called VOLTA, also known by its internal name VOLTA-AUTODOXER, that steals passwords, sessions and cryptocurrency wallets — then asks a large language model to rank what each victim is worth.

Illustration of stolen digital files being sorted into ranked folders by an automated system, no text or logos

Image: An illustration of stolen digital files being sorted into ranked folders by an automated system. Credit: Red Eye Eagle News / AI illustration.

Editor's Highlight: The shift is not the theft itself, which researchers have seen for years, but the triage: VOLTA turns a messy haul of stolen credentials into a prioritised dossier, and it keeps trying other AI models if the first one refuses.

Table of Contents

  • Key Facts
  • What VOLTA Steals First
  • How the AI Ranking Stage Works
  • A Fallback Chain Built to Expect Refusals
  • Why Researchers Think It Is Built to Be Sold
  • What Is Not Yet Known
  • How to Reduce the Risk
  • Bottom Line
  • FAQ
  • Sources

Key Facts

DetailInformation
MalwareVOLTA, internally named VOLTA-AUTODOXER
TypeSelf-deleting Windows infostealer
Reported byCisco Talos, October 31, 2026
AI stepEvidence bundle sent via OpenRouter for a target assessment
Default modelAnthropic Claude Opus 4.6, according to Talos
Fallback modelsMoonshot AI Kimi K2, then Google Gemini 2.5 Flash
Sample first seenOctober 11, 2026, at 12:46 UTC
Use in the wildNot tracked by Talos; no victim count reported

What VOLTA Steals First

According to Cisco Talos, VOLTA starts with a broad, familiar sweep. In a single pass it can take browser passwords, cookies and saved payment cards from Chromium-based browsers, along with data from Firefox and Thunderbird.

The haul goes further: Outlook profiles, Slack workspaces, Steam and Battle.net sessions, Discord tokens, Telegram sessions and Roblox accounts, plus Microsoft OAuth tokens, SSH and cloud-provider keys, Wi-Fi profiles and screenshots. Crypto wallets get special attention: VOLTA runs an embedded cracker against the wallet vault and records the result on a line labelled “WALLET VAULTS CRACKED”.

How the AI Ranking Stage Works

After the first sweep, VOLTA uses stolen sessions to enrich its picture of the victim. Using the victim's own authenticated Gmail access, it searches for purchase receipts from Amazon, DoorDash, Uber Eats, Grubhub, Walmart and PayPal to corroborate a home address, pulls food-delivery order history, and values the Roblox inventory via Rolimons.com, according to Talos.

That material is assembled locally into a structured bundle, saved as Identity_Report.txt and serialised as JSON. It is then posted to OpenRouter, a service routing requests to many model providers, under a prompt telling the model to act as “a forensic intelligence analyst drafting a target assessment”.

The requested output includes “WHAT THEY'RE WORTH”, ranking valuables by monetisation value across low, medium and high tiers, and “THEIR OPSEC”, assessing security posture and password hygiene. The answer is saved as AI_Analysis.txt and sent out with the dossier, after an AI round-trip on the victim's own machine and network.

A Fallback Chain Built to Expect Refusals

VOLTA's authors appear to have planned for a model refusing the request. Talos says the malware checks each response for refusal phrases such as “I can't” or “I'm not able” before moving to the next model in a chain of up to three.

StepModelRole
1Claude Opus 4.6Default model
2Kimi K2First fallback
3Gemini 2.5 FlashSecond fallback

Requests are hand-assembled JSON at a low temperature of 0.2, which Talos interprets as an emphasis on accuracy and consistency. Analysis Note: Talos describes the chain as implying “a direct engineering effort and expectation of hitting model guardrails” — refusal treated as an expected step, not a failure.

Talos researcher Ryan Fetterman told WIRED the prompt is “pretty carefully crafted not to imply anything overtly malicious”, with failover planned “to more permissible models” down the chain.

One limit matters: Talos does not say how any of the three models actually responded. The chain shows intent and design, not a confirmed refusal bypass. Without an API key, VOLTA skips the AI stage and still works as a conventional stealer.

Why Researchers Think It Is Built to Be Sold

VOLTA does not hardcode its Telegram channel, API keys and settings, as Talos says a tool written for personal use likely would. Instead, it uses a template with six swappable slots, filled in at build time.

Talos says this points to a builder model: the author sells access, the customer supplies their own values, and a custom binary is compiled for each order — letting “a customer of any level of sophistication” receive ranked assessments in their Telegram inbox. Every observed endpoint was third-party controlled, Talos notes, and the malware deletes itself after running.

What Is Not Yet Known

Talos is clear about the boundaries of its finding. The sample was first seen on October 11, 2026 and surfaced through its CAIRN research effort, which hunts AI-integrated malware through metadata such as embedded prompts and API keys. Fetterman told WIRED that Talos did not track whether VOLTA has been used in the wild, and no victim count, distribution method or operator identity has been reported.

DateEvent
September 22, 2026Talos releases and open-sources CAIRN
October 11, 2026VOLTA sample first seen, 12:46 UTC
October 31, 2026Talos publishes VOLTA analysis

This is therefore a documented capability and a warning, not a confirmed outbreak. Treating it as an epidemic would overstate what the evidence shows.

How to Reduce the Risk

Much of VOLTA's token theft depends on bypassing App-Bound Encryption in Chrome. Talos says an application control policy restricting which processes may launch Chrome with the --headless flag would block that decryption path entirely, and at minimum recommends alerting on unexpected headless Chrome launches.

For home users, the everyday equivalents still apply: download software only from trusted sources, be wary of unexpected downloads, and use a password manager rather than saving passwords in the browser where possible. If a machine is believed compromised, Talos advises preserving the log at %APPDATA%\volta_debug.txt, revoking compromised tokens and changing passwords.

Practical Takeaway: Any crypto wallet whose seed phrase sat in plaintext should be treated as fully compromised, with funds moved to a new wallet whose seed is generated on an air-gapped device. Changing a password alone is not enough if a live session token was stolen — the session itself must be revoked.

Bottom Line

VOLTA shows a plausible next step for crimeware: steal everything, then let a commercial AI model decide what matters most. What to watch next is whether other stealers copy the pattern, whether providers detect this kind of bulk victim-profiling request, and whether VOLTA appears in a confirmed live campaign.

FAQ

What is VOLTA?

A self-deleting Windows infostealer analysed by Cisco Talos. It steals credentials and wallets, then sends a structured bundle to a language model to rank the victim's assets.

Does the AI model steal the data?

No. The malware does the stealing, according to Talos. The model analyses already-collected data and produces a ranked assessment.

Which AI models does VOLTA try?

Claude Opus 4.6 via OpenRouter by default, Talos says, falling back to Kimi K2, then Gemini 2.5 Flash on a refusal phrase. Talos does not report how any model actually responded.

Has VOLTA attacked many people?

Unknown. Talos did not track whether VOLTA has been used in the wild, and gives no victim count or operator identity.

What should I do if I think my computer was infected?

Preserve the %APPDATA%\volta_debug.txt log, revoke compromised tokens, change passwords, and move crypto funds to a new wallet with an air-gapped seed if the old seed was stored in plaintext.

Sources

Join the conversation

Comments

Stay connected

Follow the news.

New reporting, in your feed reader. No account required.