New AI-Powered Malware 'Volta-Autodoxer' Targets Victims' Data

Imagine checking your online accounts, ordering takeout, and chatting with friends online, only to have a malicious program harvest your every digital footprint, package it into a neat dossier, and hand it to an artificial intelligence to determine how much you are worth. This is no longer speculative fiction. Security researchers have uncovered a sophisticated new strain of infostealer malware named "Volta-Autodoxer" that integrates commercial large language models to automatically profile and triage victims in real time.

Discovered by Cisco Talos researchers—who flagged the threat using their new CAIRN framework designed to track AI-integrated hacking tools—Volta represents a significant evolution in how automated malware operates. While the creators' tracking has not yet confirmed attacks in the wild, the technical implementation demonstrates how easily modern cybercriminals can leverage commercial AI to scale their operations.

src="https://media.wired.com/photos/6ae4e9f503c2c36f4cfc913c/master/w_1600,c_limit/KernelPanic_10312026_v1.jpg" alt="Photo-Illustration of Volta-Autodoxer AI Infostealer Malware" class="w-full h-auto rounded-2xl shadow-lg border border-black/10 dark:border-white/10 object-contain max-h-[550px] mx-auto block" loading="lazy" /Photo-Illustration: WIRED STAFF; Getty Images

A Wide Net Followed by Deep AI Analysis

At its core, Volta functions as a mature, aggressive infostealer designed primarily to target younger internet users with extensive digital profiles. During its initial sweep, the malware targets stored browser credentials, Microsoft OAuth tokens, cookies, active payment cards, and cryptocurrency wallet seed phrases. It also extracts session tokens from platforms including Discord, Telegram, Steam, and Roblox.

Once the initial harvest is complete, Volta performs a secondary, deeper extraction. By scanning central hubs like Gmail, the malware gathers receipts, order histories, and location data from delivery services such as UberEats and DoorDash. It compiles this vast collection of personal data into a serialized dossier.

Instead of leaving the data triage to human operators, the malware sends this evidence bundle to external AI models via OpenRouter. By default, Volta prompts Anthropic’s Claude Opus 4.6, but it also includes built-in fallbacks to Moonshot AI’s Kimi K2 and Google’s Gemini 2.5 Flash in case guardrails trigger model refusals.

Inside the 240-Word Forensic Prompt

The ingenuity of Volta lies in its carefully structured instructions. The malware embeds a 240-word prompt instructing the receiving LLM to act as a forensic intelligence analyst. According to Cisco Talos security researcher Ryan Fetterman, the prompt tells the AI to generate an executive summary detailing what the victim does, "what they’re worth," and an assessment of their operational security.

The prompt instructs the chatbot to cross-reference fields—such as checking whether a victim's Wi-Fi password matches their last name—and to focus on deep analysis rather than wide summaries. Because authentication tokens expire rapidly and passwords can be updated, automating this intelligence-gathering phase allows attackers to instantly identify and exploit high-value targets before credentials become obsolete.

The Growing Accessibility of AI-Driven Cybercrime

While low-level threat actors frequently complain about low-quality AI outputs flooding underground forums, major cybersecurity firms continue to document advanced groups utilizing large language models for everything from code generation to data analysis. Crowdstrike recently reported tracking multiple distinct hacking groups across different language regions leveraging AI to compromise global enterprises and government bodies.

Cisco Talos notes that Volta contains design clues suggesting it was built to be commercialized and sold to a broad range of malicious actors rather than kept for a single targeted campaign. Although the core payload remains a traditional infostealer that can execute without its AI component, the addition of automated LLM triage highlights an ongoing trend: advanced automation and analysis capabilities are becoming increasingly accessible to lower-tier threat actors.