New Malware Asks an AI What Each Victim Is Worth. Make Your Office a Bad Pick.
Researchers found a password stealer that hands its loot to an AI model for a ranked profile. The defense is mostly what you should already be doing, done this week.
By John Martines · November 1, 2026 · 6 min
Cisco Talos documented a password stealer that reads your email receipts, guesses your home address and asks an AI to rank what your accounts are worth. Here is what it does and the six steps that make a small office a poor target.
The owner of a small landscaping company downloads what looks like a free invoice template on a Sunday night, on the laptop she also uses for payroll. Nothing seems to happen. The file closes. She goes to bed.
By morning a stranger has a copy of every password her browser saved, the sessions that keep her signed in to email and the bank, a list of her recent online orders, and a written profile: where she likely lives, what she does for work, which accounts look most valuable, and how careful she seems to be with security.
Most of that is old news. Password stealers have done the first part for years. The written profile is new, and an AI model wrote it.
An infostealer is malware that grabs whatever a computer remembers: saved passwords, browser cookies (the small files that keep you signed in), autofill details, crypto wallets. It sends them to the attacker and often deletes itself. A stolen session can be worse than a stolen password, because it may let someone into your account without the password or the code on your phone.
On October 31, 2026, Cisco's threat research team, Talos, published an analysis of one called Volta, nicknamed the "autodoxer." "Doxing" means pulling together someone's private details. Talos found that Volta goes further than most stealers:
- It uses the stolen sessions to search Gmail for store receipts.
- It checks food delivery order history, such as DoorDash and Uber Eats, to confirm a home address.
- It packs everything into a bundle and, if the attacker has set it up, sends the bundle to an AI model with instructions to write an assessment: where the person lives, what they do, what their accounts are worth ranked low, medium and high, and how good their security habits look.
It was set to use a Claude model through a service that routes requests to many AI models, and if a model refused, it tried others, up to three. The AI step is optional. Without it, Volta still steals everything.
The important caveat: Talos told WIRED it had not tracked whether Volta has been used against real victims. Treat it as a working example of where these tools are heading, not proof that AI is already sorting thousands of victims a day.
How it got here:
- For years: infostealers have been a common first step in account takeovers. Stolen logins and sessions are sold in bulk to other criminals.
- March 6, 2026: Microsoft's threat intelligence team reported that state-linked hacking groups use AI to write phishing messages in several languages and to tailor lures from information scraped about their targets.
- October 31, 2026: Cisco Talos published its Volta analysis, and WIRED reported it the same day.
A stealer that collects ten thousand logins has a sorting problem: which ones are worth a criminal's time? An AI profile solves that. It reads the pile and points at the person with payroll access, the supplier portal and the business bank account.
That person is usually you. In a small business the owner or office manager often holds the keys to everything, on one laptop, in one browser, mixed with personal shopping and email. That is exactly the profile a ranking would put near the top.
Three things follow.
One laptop can open the whole business. If email, bank, payroll and software accounts all live in the same browser, one bad download hands over all of them.
The follow-up gets personal. A profile that knows your suppliers, your recent orders and your address makes the next scam convincing: an invoice from a real vendor, a delivery problem with a real order, a call that sounds like someone you know.
A strong password isn't enough. A stolen session can skip both the password and the sign-in code. You need to be able to end sessions, not just change passwords.
1. Make downloads the narrow door. Most stealers arrive as something you chose to install: a free template, a cracked program, a fake AI app, a "browser update" from an ad. Take away administrator rights on staff computers so new software needs a yes from your IT partner, and install only from vendor sites you typed in yourself. [Fake AI app downloads](/articles/fake-ai-app-downloads) are one of the most common lures right now.
2. Separate work from home in the browser. Use a separate browser profile, or better, a separate computer, for the business bank, payroll and admin accounts. Keep personal shopping, personal email and delivery apps out of it. A stealer can only take what that browser holds.
3. Stop saving passwords in the browser. Use a password manager with its own lock, and turn off "save password" in the browsers on work machines. It won't stop every stealer, but it removes the easiest pile.
4. Learn the "sign out everywhere" button. Microsoft 365, Google Workspace, your bank and your payroll service all show where you're signed in and let you end every session. Find it now, while nothing is wrong. If a laptop ever acts strangely, end all sessions first, then change passwords from a different, clean device.
5. Keep the protection current. Make sure every company computer has security software that your IT partner actually watches, with updates on. Ask them to confirm it in writing, computer by computer.
6. Keep the call-back rule. Any change to a payment, bank detail or vendor address gets confirmed by calling a number you already had, never one in the email. That one rule stops most of what a stolen profile is used for.
- Which of our computers let staff install software without approval?
- Do we have security software on every machine, and who sees its alerts?
- Can you sign out every session for one person across our accounts, and how fast?
- Are passwords saved in browsers on our work machines today?
- If an infostealer hit the owner's laptop tonight, what would you do in the first hour?
Criminals have always had more stolen data than time. AI is giving them a cheap way to sort it, and to write the follow-up message once they have picked a target. Security researchers are also using AI to spot malware that calls AI models, so the contest runs both ways.
You can't control whether your name lands in a pile. You can make the profile boring: few saved secrets, a clean split between work and home, sessions you can cut off in minutes, and a staff that calls before paying. That is the kind of target a ranking puts at the bottom.
---
Sources (checked November 1, 2026). Volta's behavior, the Gmail receipt search, delivery app order history, the optional AI stage, model fallback and the ranking prompt: Cisco Talos, "Appraisal by AI: Model-driven victim prioritization in the VOLTA infostealer" (Talos Intelligence blog, Oct 31, 2026). Not yet tracked in real attacks, and the "autodox" framing: "Spooky New Infostealer Malware Uses AI to 'Autodox' Victims" (WIRED, Oct 31, 2026). Further reporting on the analysis: "Malware called Volta-Autodoxer uses AI to build a dossier on its victims" (Startup Fortune, Oct 31, 2026) and Data Studios, "VOLTA infostealer uses Claude, Kimi and Gemini via OpenRouter to rank victims, Cisco Talos says" (Oct 2026). State-linked groups using AI for phishing: Microsoft Threat Intelligence, "AI as tradecraft: How threat actors operationalize AI" (Microsoft Security Blog, Mar 6, 2026).