A New Approach to AI-Assisted Cybercrime

Cybersecurity researchers have identified a Windows malware strain called VOLTA-AUTODOXER, also known as VOLTA, that uses artificial intelligence to assess the financial value of stolen personal information.

Unlike conventional infostealers that primarily collect passwords, payment details and account credentials, VOLTA adds an AI-powered profiling stage. It gathers information from an infected computer, builds a detailed dossier about the user and sends the data to large language models (LLMs) to identify potentially valuable targets.

Cisco Talos researchers discovered the malware through their CAIRN research initiative, which investigates malicious tools incorporating AI capabilities.

The discovery highlights how cybercriminals can use commercial AI services to automate tasks that would otherwise require manual investigation.

What Information Does VOLTA Steal?

VOLTA collects a broad range of sensitive information from infected Windows devices, including:

  • Browser passwords, cookies and saved payment cards.

  • Cryptocurrency wallet seed phrases.

  • Microsoft OAuth tokens and cloud-provider credentials.

  • SSH keys and Wi-Fi passwords.

  • Outlook profiles and Slack workspace information.

  • Discord, Telegram, Steam and Roblox session data.

  • Screenshots and document authorship metadata.

The malware also attempts to obtain additional information through accounts it has already compromised.

For example, it searches Gmail for shopping and delivery receipts from services such as Amazon, DoorDash, Uber Eats, Grubhub, Walmart and PayPal. These records can help confirm a victim's residential address. It also retrieves order histories and other account information to build a more complete profile.

Claude, Kimi and Gemini Used for Victim Assessment

VOLTA's distinguishing feature is its use of commercial AI models to analyse the collected information.

The malware sends a structured evidence bundle through OpenRouter, a service that provides access to multiple AI models.

According to Cisco Talos, its model configuration uses:

  • Claude Opus 4.6: The default model from Anthropic.

  • Kimi K2: A fallback model from Moonshot AI.

  • Gemini 2.5 Flash: Google's model used as another fallback.

The malware checks AI responses for refusal phrases such as "I can't" and "I'm not able". If a model refuses, it can move to another model in the chain.

The AI is instructed to act as a forensic intelligence analyst and produce a target assessment. One section, titled "WHAT THEY'RE WORTH", ranks the victim's assets according to their potential monetisation value. Another, titled "THEIR OPSEC", evaluates password hygiene and security practices.

This allows attackers to move beyond collecting stolen information and automatically identify which compromised accounts or assets may be worth pursuing.

Why AI Makes Infostealers More Concerning

Traditional infostealers already create serious risks by exposing credentials and authentication sessions. VOLTA adds another layer by reducing the manual effort needed to interpret stolen information.

Cybercriminals often have limited time to exploit stolen credentials and session tokens before victims change passwords or revoke access. Automatically identifying potentially valuable accounts could help attackers prioritise their next actions.

Cisco Talos describes this development as an example of AI acting as an intelligence analyst for the malware operator.

Researchers also believe VOLTA may have been designed as a customisable malware builder, allowing different customers to supply their own Telegram channels and API credentials.

Importantly, the AI component is optional. VOLTA can still operate as an infostealer even when the AI analysis stage is unavailable.

How Individuals and Businesses Can Reduce Risk

The discovery reinforces the importance of protecting credentials, authentication sessions and personal information.

Individuals should avoid downloading cracked software, unofficial applications and suspicious attachments, as these can be used to distribute infostealers.

Businesses should strengthen endpoint protection, monitor unusual browser activity and enforce multi-factor authentication. Organisations should also review exposed credentials and revoke compromised sessions rather than relying only on password changes.

If a device is suspected of infection, affected users should change passwords from a clean device, revoke active sessions and rotate exposed API keys. Cryptocurrency wallets whose seed phrases were stored on the compromised device should be treated as compromised.

The Growing Intersection of AI and Cybercrime

VOLTA demonstrates how existing malware can become more capable by integrating commercial AI services into its operations.

The technology is not independently carrying out every stage of a cyberattack. Instead, it is automating the analysis and prioritisation of information already stolen by the malware.

As AI tools become more accessible, cybersecurity teams may face threats that combine traditional credential theft with automated profiling and decision-making.

The challenge will be to detect not only the theft of sensitive information but also how attackers use that information after a compromise.