VOLTA Malware Uses LLMs to Rank Victims by Wealth

What VOLTA actually does

VOLTA is a self-deleting Windows infostealer that harvests a wide range of credentials and sessions, then uses an LLM to turn that raw data into a prioritized target assessment.

The malware grabs browser passwords, cookies, payment cards, cryptocurrency wallet seed phrases, Discord and Telegram sessions, Steam and Roblox accounts, Microsoft OAuth tokens, SSH and cloud-provider keys, Wi-Fi profiles, and screenshots. It also runs an embedded cracker against crypto wallet vaults and dumps the results.

But the real novelty is what happens next. VOLTA assembles a structured evidence bundle about the person behind the machine—including geolocation, employer, inferred birth year, card last-4 digits, Slack workspaces, document authorship metadata, and even a live Roblox account valuation—and sends it to an LLM with a system prompt instructing the model to act as a forensic intelligence analyst.

The model is asked to produce a section titled “WHAT THEY'RE WORTH” with valuables ranked by monetization potential, plus an assessment of the victim's operational security and password hygiene. The output is a citation-backed dossier that turns a messy credential log into a clear list of what to steal first.

The multi-model fallback

VOLTA anticipates that some models will refuse to help with criminal profiling, so it checks each response for refusal phrases and moves on to the next model in a chain of up to three.

The malware uses OpenRouter's API to access multiple LLMs. It sends a hand-assembled JSON payload with a low temperature of 0.2, which suggests the operator wants consistent, accurate output rather than creative responses.

If a model replies with something like “I can't” or “I'm not able,” VOLTA tries the next model. The fallback chain likely includes smaller, more permissive models that are less likely to refuse. This engineering effort shows the operator expects to hit guardrails and has planned around them.

The entire inference round-trip happens on the victim's own machine, over their own network connection, before the results are exfiltrated in a zip file. That means the attacker doesn't need to maintain infrastructure for the AI processing—it's all done locally and then sent out.

Live enrichment: Gmail, Uber Eats, and Roblox

VOLTA doesn't just steal data—it uses stolen sessions to query live services and enrich the victim's profile before the LLM analysis.

Using the victim's authenticated Gmail tokens, VOLTA searches the inbox for receipts from Amazon, DoorDash, Uber Eats, Grubhub, Walmart, and PayPal to confirm a home address. It also pulls order history directly from Uber Eats and DoorDash sessions, enriches Discord profiles with stolen tokens, and values Roblox inventories via a live lookup on Rolimons.com.

This enrichment turns a static credential dump into a dynamic, up-to-date picture of the victim's life and assets. The LLM then connects the dots—for example, linking names, addresses, or universities to patterns in passwords—to stretch the data further than a human analyst could manually.

What it means for you

VOLTA is a Windows infostealer, so it primarily affects Windows users. There is no indication it targets macOS or Linux.

The malware is distributed through unknown channels, but infostealers typically spread via phishing emails, malicious downloads, and cracked software. Because VOLTA self-deletes after exfiltration, it can be hard to detect after the fact.

The best defense is standard: keep Windows and browsers updated, use a password manager, enable two-factor authentication everywhere, and avoid downloading software from untrusted sources. If you suspect an infection, change your passwords from a clean device and revoke active sessions. For a deeper look at how AI agents are changing the threat landscape, see our coverage of enterprise AI agents making risky security decisions.

Availability and impact

VOLTA is not a commercial product; it's malware observed in the wild. There is no official release or pricing. The story is relevant to Windows users worldwide, particularly in the US, UK, and EU, where infostealers are a common threat.

No specific campaigns targeting these regions have been disclosed, but the techniques described are broadly applicable. Users should remain vigilant regardless of location.

VOLTA shows how AI can be weaponized not just to automate attacks but to add a layer of intelligence. As LLMs become more accessible, expect more malware to incorporate similar profiling capabilities. The arms race is heating up.