-
A new Windows infostealer going by the name VOLTA has emerged. It uses AI models to analyze stolen data and rank each victim by profit potential.
-
The malware can read Gmail receipts to confirm home addresses before sending a target report to the hacker on Telegram.
-
Security researchers at Cisco Talos have studied one sample of the infostealer, but they’ve not tracked real attacks, so the true scope remains unknown.
Criminals now have an AI assistant to sort their loot. Cisco Talos researchers say a new Windows malware called VOLTA hands stolen data to AI models. The models then rank each victim by profit potential.
Talos published its findings on October 31. The malware calls itself VOLTA-AUTODOXER. It uses a technique called ‘doxing’ to collect someone’s private details, and VOLTA happens to automate the job.
What information does VOLTA steal?
VOLTA starts like any other infostealer. It goes after saved passwords and login cookies from browsers like Chrome. Its other targets include payment cards and Wi-Fi passwords, and it takes screenshots as well. In addition, it goes after crypto wallets. It copies seed phrases and also tries to crack wallet vaults on the spot.
But that’s not all. VOLTA grabs sessions from Discord, Telegram, Steam, and Roblox. Microsoft sign-in tokens, Slack data, and Outlook profiles aren’t safe either. On top of that, it pulls SSH and cloud keys those are the keys that unlock servers and cloud accounts.
Stolen data alone doesn’t reveal much about who someone is. So VOLTA digs deeper, it uses the victim’s logged-in Gmail to search around for receipts. It’ll scan for emails from platforms such as Amazon, DoorDash, Uber Eats, Grubhub, Walmart, even PayPal. These receipts often have people’s home addresses.
Additionally, it digs up order histories from Uber Eats and DoorDash. It also uses a stolen Discord token to fill in profile details. It even puts a dollar value on Roblox items.
The result is a file on the person behind the computer. Talos says it can include a home address, employer, estimated birth year, and the last four digits of payment cards. It can also list Slack workspaces and signs of iPhone backups.
How the AI ranks victims
This is where it gets interesting. VOLTA transmits the file to OpenRouter, a service that directs queries to a variety of AI firms. By default, it submits the inquiry to Anthropic’s Claude Opus 4.6.
It has a pre-programmed prompt that instructs the model to behave like a forensic intelligence analyst. This prompt has a length of approximately 240 words. Further, it directs the model to use just the facts contained within the file.
The model must write two sections. The first one, called “What They’re Worth,” separates valuables into low, medium, or high categories. The second, “Their OPSEC,” reviews the victim’s security habits and password strength.
The author apparently saw pushback coming. So, they prepared for it VOLTA checks every answer for refusal phrases like “I can’t.” If it spots one, it switches over to Moonshot AI’s Kimi K2. If that doesn’t work, it goes for Google’s Gemini 2.5 Flash. Talos looks at this sequence as a strategy to handle AI safety limits. Still, Talos never said that any of these models actually wrote a report.
How the malware gets the data across to the attacker
VOLTA sends data to the attacker over Telegram in three steps. The first batch leaves right after the theft. It holds passwords, cookies, cards, wallets, and screenshots.
The second batch adds the AI analysis. The third adds the Gmail receipts and order histories. Talos describes VOLTA as a self-deleting malware, meaning it deletes itself from the system after completing the task.
The malware uses a template that has six empty slots. They hold items such as the Telegram channel and API keys. Talos believes the author either sells a builder kit or makes custom copies on request. In either case, a buyer needs little skill. The ranked report simply lands in a Telegram inbox.
Talos researcher Ryan Fetterman says the AI adds no new speed or scale. Instead, it brings “effort displacement.” In plain terms, the AI does the analyst’s homework, so the criminal can skip it.
Talos first saw the sample on October 11 and studied only that one. It has not tracked whether criminals use VOLTA against real people. No victim count exists. The AI step is also optional.
Without an API key, VOLTA still steals everything and skips the report. In addition, all six shops it checks are popular U.S. brands. Even so, stolen logins and wallets are valuable anywhere.
Tips to protect yourself
At its core, VOLTA is a regular Windows stealer. Therefore, the usual defenses still apply.
- Keeping your security software updated helps.
- Never keep seed phrases in plain files or screenshots.
- Use a password manager; don’t just save passwords in your browser.
- Also turn on two-factor sign-in for email and crypto accounts.
However, two-factor authentication alone cannot stop every AI-powered attack. Google has also reported cases where hackers used AI tools to plan large-scale attacks designed to bypass two-factor authentication.
The AI request leaves from the victim’s own computer and internet connection. Therefore, companies can watch for odd outbound traffic to openrouter.ai. Talos flagged that traffic as a clue. Home users should scan their PCs and change passwords if they spot unknown logins.
This is not the only AI-powered threat. Last week, Talos described a separate tool called NIK C2. It can borrow a victim’s own OpenAI key to run its analysis. Criminals, it seems, are quickly adding AI to old tricks.
