VOLTA malware enlists AI to rank victims’ data by financial value

How VOLTA works

Cisco Talos researcher Ryan Fetterman published analysis of VOLTA-AUTODOXER on October 31, 2026, following its discovery through the CAIRN threat-research program. First observed on October 11, the malware combines traditional credential theft with an AI-powered victim-scoring stage that sets it apart from conventional infostealers.

After infiltrating a Windows machine, VOLTA harvests a broad range of data: browser passwords, cookies, and saved payment cards from Chromium and Firefox; email credentials from Thunderbird and Outlook; Slack workspace sessions; chat data from Discord and Telegram; gaming accounts on Steam, Battle.net, and Roblox; cryptocurrency wallet seed phrases; Microsoft OAuth tokens; SSH and cloud provider keys; Wi-Fi profiles; and screenshots.

AI models as the sorting layer

Once data is exfiltrated, VOLTA routes it through OpenRouter — a commercial API gateway for large language models — and submits it to AI models including Anthropic’s Claude, Google’s Gemini, and Moonshot AI’s Kimi. The models return a victim dossier ranking stolen data by monetization potential alongside an assessment of the target’s security posture.

If one model declines the request, VOLTA automatically retries with up to two alternatives before falling back to operating without the AI stage. The AI layer is optional: operators without an API key still receive raw stolen data.

No attacker infrastructure to target

VOLTA deletes itself after execution and communicates only through third-party services — the AI APIs and OpenRouter — rather than attacker-controlled servers. This design makes network-level blocking and infrastructure takedowns difficult.

What to do

Talos has not documented any large-scale VOLTA campaigns as of publication. Even so, the malware signals a broader trend: AI is being embedded in criminal toolchains not to generate attacks but to maximize return on stolen data.

Talos recommends restricting which processes can launch Chromium with the headless flag, alerting on unexpected headless browser launches, and storing sensitive credentials in hardware-backed credential stores where possible.