An infostealer can collect a person’s browser data faster than an operator can make sense of it. Cisco Talos’s new VOLTA research examines malware that hands that second job to a language model.
In its October 31 technical report, Talos describes a Windows infostealer that optionally turns harvested credentials and personal records into a victim profile. The model ranks assets by their potential value to the operator.
Stolen sessions can enrich the model’s input with account information and purchase history. The mechanism connects access to an account with clues about its owner. For a company investigating exposed work credentials, that creates a reason to examine the surrounding personal-to-work connections as well as the credential itself.
Talos documents requests through OpenRouter and a fallback across Claude, Kimi and Gemini models when responses are refused. The stealer also works without AI. A refusal from one provider leaves other routes and the underlying theft to consider.
WIRED’s accompanying report describes the AI stage as a way to surface valuable targets quickly. It also supplies a firm limit: Talos did not track whether VOLTA had been used in the wild. An analyzed capability establishes neither a victim count nor a measured increase in successful attacks.
For defenders, Talos recommends preserving the malware’s harvest log when available and revoking potentially compromised tokens as well as changing passwords. Those records can help scope what was collected and whether the model call succeeded.
An enterprise response should follow the exposed sessions through to the systems they reach. Cleaning the machine addresses the implant; someone still has to account for the access and personal information that may already have left it.