VOLTA-AUTODOXER infostealer uses AI to rank victims for theft
A new Windows infostealer named VOLTA-AUTODOXER leverages large language models to automatically profile victims and rank stolen data by financial value.

Cybersecurity researchers have identified a new Windows infostealer called VOLTA-AUTODOXER that uses generative AI to automate victim profiling and data exfiltration. The malware, commonly shortened to VOLTA, harvests a wide array of sensitive information and uses AI to assess a victim's financial worth and security posture before sending the analysis via Telegram.
VOLTA-AUTODOXER is a self-deleting infostealer that steals browser passwords, cookies, saved payment cards, and cryptocurrency wallet seed phrases. It also targets email profiles, Microsoft OAuth tokens, SSH and cloud configuration files, plaintext Wi-Fi passwords, and sessions from apps like Discord, Telegram, Steam, and Roblox. After infection, the main executable may delete itself, leaving behind a debug log file. The malware is likely a commoditized builder kit or bespoke tool operated by its author, designed for resale with six swappable template slots for custom Telegram channels and API keys.
AI Integration and Model Fallback
The malware's AI layer reduces operator effort by automating analytical work rather than improving speed or scale. Collected data is assembled locally into a file called Identity_Report.txt, serialized to JSON, and sent via a POST request to the OpenRouter API with a detailed system prompt. The AI is instructed to produce a concise assessment of the victim's suitability as a target, including an executive summary and an analysis of their operational security.
VOLTA employs a multi-model architecture with a default and fallback sequence. The inference call occurs over the victim's own network connection.
| AI Model | Provider | Purpose |
|---|---|---|
| Claude Opus 4.6 | Anthropic | Default model for analysis |
| Kimi K2 | Moonshot AI | First fallback option |
| Gemini 2.5 Flash | Second fallback option |
If the primary model refuses the request or provides a thin reply, the malware automatically switches to the next model in the chain. The 240-word prompt is designed to avoid overtly malicious language to reduce the chance of model refusal. VOLTA can operate without AI, though the analysis is an added capability. If no OpenRouter API key is provided, the stolen dossier can still be sent to Telegram without the AI output.
Exfiltration Mechanism
Exfiltration occurs via Telegram in three distinct stages: after the initial data collection, after the AI analysis call, and after a final live enrichment phase. The AI's reply is saved as AI_Analysis.txt, and both the original report and the analysis are bundled into a compressed archive named volta_dox.zip for exfiltration. This process creates a structured dossier on the victim that can include geolocation, home address, employer, inferred birth year, and the last four digits of payment cards.
Ryan Fetterman, a security researcher at Cisco Talos, highlighted the tactical shift. "Identifying high-value targets faster increases the chance of using stolen credentials before they expire or are changed," he said. The malware's AI expansion foreshadows more invasive activity in the future. Adversarial adoption of generative AI for tasks like examining massive data troves for classification is becoming the new normal in cyberattacks.
Data Collection and Enrichment
VOLTA performs a second, deeper sweep using hijacked session data. It reuses stolen access to search a victim's Gmail for receipts from services like Amazon, DoorDash, Uber Eats, Grubhub, Walmart, and PayPal. It pulls order histories from food delivery services to corroborate home addresses and build a fuller picture of the victim's location and habits. The malware also performs live enrichment by valuing stolen Roblox accounts through the Rolimons website. The final AI output includes a section ranking valuables by monetization tier and another assessing the victim's security posture.
The malware bypasses Chrome's App-Bound Encryption by launching Chrome in a headless mode. However, application-control policies that limit headless Chrome launches can reduce exposure to this specific decryption method, though they do not affect other theft, AI, or exfiltration functions.
Defensive Recommendations
Defenders are advised to assume comprehensive compromise if this malware is detected. All API keys and OAuth tokens should be revoked and rotated immediately. Passwords must be changed, not only for compromised systems but for all accounts where credentials were potentially reused. For cryptocurrency, funds should be moved from any wallet where the seed phrase was stored in plaintext to a new, air-gapped wallet. This follows the discovery of other AI-enabled threats, like the NIK C2 framework, which can abuse a victim's own OpenAI API key to fund malicious inference costs in a technique called LLMjacking.
The integration of AI marks a shift toward automated, high-value victim targeting. It increases the speed and potential impact of credential theft and fraud. Researchers at CrowdStrike have already found four distinct hacking groups using AI tools to compromise targets in government, healthcare, retail, and travel sectors. Experts recommend revoking or rotating all API keys and tokens, changing passwords, and moving cryptocurrency funds from wallets with exposed seed phrases to new air-gapped wallets.





