New malware uses AI to rank stolen data by financial value

Automating the heist triage

Cybersecurity researchers have identified a new Windows infostealer named VOLTA-AUTODOXER, which integrates large language models (LLMs) to evaluate the value of stolen information. Unlike traditional malware that simply dumps files, VOLTA acts as an automated analyst for cybercriminals. It organizes stolen data into a structured bundle to determine which victims are worth the most effort to exploit.

The malware is designed to harvest a wide array of sensitive credentials, including passwords and cookies from Chromium-based browsers. It also targets data from Outlook profiles, Slack workspaces, and payment cards. Furthermore, it captures cryptocurrency wallet seed phrases, session tokens for platforms like Discord and Roblox, and even Wi-Fi profiles from the infected machine.

According to Cisco Talos researcher Ryan Fetterman, the malware builds a comprehensive profile of the user. This bundle includes geolocation data, employer information, inferred birth years, and even a live valuation of the victim's Roblox account. By collecting these specific details, the malware provides a clear picture of the person behind the computer rather than just a list of files.

How the AI analysis works

To make the stolen data more useful, VOLTA performs a live enrichment step by searching the victim's Gmail account. It specifically looks for receipts from popular services like Amazon, DoorDash, Uber Eats, Grubhub, Walmart, and PayPal. This allows the malware to corroborate a victim's home address and gather more evidence of their spending habits.

Once the data is gathered, the malware sends the information to an AI system via OpenRouter, a unified API gateway. The malware is programmed to use Anthropic's Claude Opus 4.6 as its primary model. If that model refuses the request, the malware automatically falls back to Moonshot AI's Kimi K2 or Google's Gemini 2.5 Flash.

The malware uses a specific 240-word system prompt to guide the AI's behavior. It instructs the model to assume the role of a forensic intelligence analyst. The prompt explicitly tells the AI that the provided bundle is the only permissible source of facts, preventing the model from using its own training knowledge to hallucinate details about the victim.

Creating a prioritized dossier

The AI's primary job is to turn raw logs into a prioritized dossier for the attacker. It generates a specific section titled 'WHAT THEY'RE WORTH,' where it ranks the victim's valuables by their monetization potential. These items are categorized into low, medium, or high tiers based on how easily they can be turned into profit.

In addition to the financial ranking, the AI provides an assessment of the victim's security posture. This section, titled 'THEIR OPSEC,' evaluates the victim's password hygiene and overall security habits. This allows the attacker to understand how well-protected the target is and where the next vulnerabilities might lie.

The malware exfiltrates this information to the attacker via Telegram in three distinct stages. The first stage happens immediately after the initial theft of passwords and cookies. The second stage occurs after the AI call, which includes the analysis results. The final stage happens after the live enrichment step, which adds the Gmail receipts and order histories.

A tool for the masses

Researchers believe that VOLTA might be a commoditized tool, meaning it could be sold to various cybercriminals. The malware includes a template with six slots that are filled in at the time of build, such as the Telegram channel for exfiltration and the necessary API keys. This suggests the author may be selling a builder kit or creating custom binaries for specific customers.

If VOLTA is indeed a commercial product, it means that any attacker, regardless of their technical skill, can use it. A customer can simply point the tool at a victim and receive a ranked target assessment in their Telegram inbox. This effectively removes the need for the attacker to perform manual analysis on the stolen data.

The primary goal of adding AI to this malware is not to increase the speed of the attack. Instead, the goal is effort displacement. By using AI to handle the tedious work of sorting through data, the malware allows the operator to focus on the actual exploitation of the high-value targets identified by the system.

The rise of AI in crime

The emergence of VOLTA is part of a broader trend where threat actors are adopting generative AI to improve their workflows. Attackers are increasingly using these technologies to develop custom tools and find sensitive information more efficiently. This shift increases the likelihood of success for cyberattacks across various industries.

Security experts note that generative AI is particularly well-suited for examining massive amounts of data for classification or content searching. Once an attacker gains access to a system, they can use AI to scan every accessible file to find the most sensitive data. This can be used to maximize the pressure on a victim to pay a ransom during a ransomware attack.

Other examples of AI-enabled malware have already appeared, such as the NIK C2 framework. That tool allows attackers to use a victim's own OpenAI API key to fund the AI inference. This technique, known as LLMjacking, allows bad actors to run their operations while the legitimate owner pays for the computing power used to analyze the stolen data.