A new infostealer called Volta-Autodoxer adds AI analysis to a familiar form of cybercrime: stealing credentials and personal data from infected devices. The software can collect information from accounts and apps, then send it to language models to help attackers assess which victims may be most valuable.
From stolen credentials to a victim profile
Infostealers are often hidden in counterfeit software. They can take usernames, passwords, authentication tokens, cookies and payment card data from infected machines. Stolen credentials have also been used to access targets in major data breaches, and infostealers have taken hundreds of millions of credentials in recent years.
Volta casts a particularly broad net. It searches for browser passwords, Microsoft OAuth tokens, cookies, payment cards and crypto wallet seed phrases. It also looks for active Discord, Telegram, Steam and Roblox sessions.
After its initial sweep, Volta can use what it has found to search more deeply. Access to Gmail or another central account may reveal receipts and order histories from services such as UberEats and DoorDash, helping the malware assemble details about a victim’s location and habits.
AI turns collected data into an assessment
The malware packages its findings into a dossier and sends them to AI models through OpenRouter, a service that routes prompts to different AI providers. Its default model is Anthropic’s Claude Opus 4.6, with Moonshot AI Kimi K2 and Google’s Gemini 2.5 Flash listed as fallbacks.
The instructions ask the model to act as a forensic intelligence analyst and produce a structured assessment. The requested profile includes an executive summary, what the person does, an estimate of “what they’re worth,” and a review of “their opsec.” The prompt also encourages the model to connect details across the collected information.
That analysis could help attackers sort victims and decide where to focus. Cisco Talos researcher Ryan Fetterman described the AI component as a prioritization and analysis function: the malware collects broadly, then uses a model to interpret the resulting dossier.
AI is an addition, not a requirement
Volta’s AI calls are optional. Fetterman says the malware can still operate as an infostealer without them, so the model analysis adds capability rather than enabling the basic theft.
The researchers say the prompt is written without making an overtly malicious request. Volta also has a chain of alternative models in case one refuses to respond. The source describes fallbacks and the use of cost-effective, older-generation models as common techniques among attackers.
Cisco Talos identified Volta using its CAIRN framework, which is designed to detect hacking tools with AI capabilities. The researchers did not track whether Volta has been used in the wild, so its presence in the research does not establish that victims have been targeted with it.
A wider shift in how stolen data is handled
Volta is one example of hackers applying language models to tasks around cybercrime. Security researchers have also reported criminals using AI for coding and for analyzing large collections of stolen information. In a separate account cited by the source, Crowdstrike researchers described four hacking groups using AI tools to compromise targets around the world, including government entities and businesses in health care, retail and travel.
Fetterman says clues in Volta suggest it may have been designed for sale to a range of attackers, rather than for one actor’s campaigns. If profiling helps attackers identify promising targets faster, they may have more opportunity to use credentials or tokens before passwords change or tokens expire. The broader concern is that AI-assisted analysis can make a traditional infostealer more useful to attackers with varying levels of sophistication.