New Data-Stealing Malware Uses AI to Profile Its Victims
Cisco Talos researchers have found Volta-Autodoxer, a malware strain that sends stolen data to AI language models to automatically assess victims and pick out the most valuable ones. It is not known whether it has been used in real attacks.
Researchers at cybersecurity firm Cisco Talos have discovered a new type of data-stealing malware, known as an infostealer, that uses AI language models to automatically build profiles of infected people and highlight the most valuable targets. The finding was first reported by Kernel Panic, a newsletter from Wired. The researchers named the program Volta-Autodoxer because that is how it refers to itself internally. It shows how commercial language models are being used to enhance established kinds of malware.
What it steals
Infostealers are often hidden in counterfeit software and grab usernames, passwords, authentication tokens, cookies and payment card data. Many major breaches in recent years, including the Snowflake-linked incidents that affected AT&T and Ticketmaster and the theft of a terabyte of Disney data, resulted from attackers using login details collected this way.
Volta first gathers passwords saved in browsers, Microsoft OAuth tokens, cookies, payment cards and crypto wallet seed phrases, as well as active Discord, Telegram, Steam and Roblox sessions. It then runs a second, deeper sweep, for example using a Gmail account to learn about the victim's location and habits from UberEats and DoorDash receipts and order history. According to Talos researcher Ryan Fetterman, the program appears geared toward younger internet users with active digital lives.
How AI is used
The collected data is sent to language models through OpenRouter, a service that routes prompts to several AI providers. By default Volta uses Anthropic's Claude Opus 4.6, with fallbacks to Chinese developer Moonshot AI's Kimi K2 and then Google's Gemini 2.5 Flash. The model is asked to produce a summary of the person, what they do, "what they're worth" and their security habits. Fetterman notes the malware still works without the AI component.
Talos did not track whether Volta has been used in the wild. The researcher believes it may have been built to be sold to various attackers. He also warns that identifying high-value targets faster raises the chance of using stolen credentials before tokens expire or passwords are changed.
