Artificial intelligence is advancing cybercriminals' ability to gather and analyze personal information from food delivery orders and online shopping receipts, according to research released this month by Cisco Talos.
The Windows-based malware called “VOLTA-AUTODOXER” is augmented by a large language model that sorts through passwords, payment cards, Wifi profiles and conversations on Discord, Telegram and Roblox and builds a victim profile that can include geolocation, home address, employer and inferred birth year.
Ryan Fetterman, a security and threat researcher at Cisco’s threat intelligence group Talos, said the technological capability “feels like a ‘Black Mirror’ episode.”
“The distinguishing factor for me is that level of AI integration
that maybe makes this more effective to deploy at scale,” Fetterman told USA TODAY.
He discovered the malware in October while studying the progression of malware samples that integrate AI. Online scams have been on the rise this year, increasing by 41.1% in the past year as cybercriminals increasingly target employees with personalized scams, according to a 2026 Phishing-Threat Trends Report. Fetterman said this is the first time AI has been used to augment malware in this way.
VOLTA is a Windows-based “stealer malware,” meaning it is a type of discrete software designed to extract sensitive data from an infected Window’s device without the user’s knowledge. It primarily targets the browser and email providers like Outlook and Thunderbird, looking for stored credentials.
“It does a pretty wide pass on like this initial set of secrets that it can collect from you” Fetterman said, before it does a second pass to leverage access to authentication tokens, which are digital credentials it can use to pretend to be an individual to gain access to accounts.
“If it has your Gmail token, it can run queries against your Gmail inbox to pull that second level of information, and that's where the things like your Uber Eats or your DoorDash receipts or your Amazon Prime deliveries, that kind of context, can be pulled out of your Gmail inbox,” he added.
After the malware pulls that context, it organizes the personal information into a cohesive bundle and posts it on OpenRouter, an AI model marketplace where different AI models can run the same task, acting as a backup if one model fails.
While it works without AI, using an optional AI integration prompts the model to guess what a potential victim is worth and assess security posture and password hygiene.
“The cyber use is really fully on the criminal element side,” Fetterman said. “It’s asking to use the model as a contractor in their cybercrime business.”
The information the malware finds in email or browser is organized by the AI integration to succinctly point bad actors in the direction of more vulnerable victims, pointing to how much disposable income users have and how safe their online accounts are.
“I think that's a really, probably effective and devious way to use AI because these models are really good at making disparate connections between data points,” Fetterman said. For example, it encourages connections between the victims' names, addresses or universities to patterns in their passwords.
The AI prompt used in the malware “just comes off as really consciously, carefully written to not sound like an overtly malicious request,” he added.
To prevent becoming a victim, Fetterman recommended that individuals using Windows devices shouldn’t click on random links or attachments and should turn on multi-factor authentication. Since it is still unknown how the malware is being deployed at scale, Cisco couldn’t say how to prevent it at scale
“All of the basic security best practices are still the right answer here,” Fetterman said. “It's just kind of a more interesting and disturbing development.”
Greta Reich covers the artificial intelligence industry for USA TODAY through a fellowship from the Tarbell Center for AI Journalism. Funders do not provide editorial input.
This article originally appeared on USA TODAY: Food delivery habits are giving cybercriminals personal info
