Cybersecurity researchers have uncovered **VOLTA-AUTODOXER**, a sophisticated Windows infostealer that distinguishes itself by employing a large language model (LLM) to assess and rank a victim's compromised data based on its potential monetary value. ### Comprehensive Data Harvesting Like many contemporary stealers, **VOLTA** is designed to exfiltrate a broad spectrum of sensitive information. This includes passwords and cookies from Chromium-based browsers, data from **Outlook** profiles and **Slack** workspaces, payment card details, cryptocurrency wallet seed phrases, and session tokens for platforms like **Discord**, **Telegram**, **Steam**, and **Roblox**. It also targets **Microsoft OAuth** tokens, SSH and cloud-provider keys, Wi-Fi profiles, and takes screenshots. **Cisco Talos** researcher Ryan Fetterman noted that "VOLTA builds a structured evidence bundle about the person behind the machine." This bundle encompasses geolocation, home address, employer details, inferred birth year, last four digits of payment cards, Slack workspaces, document authorship metadata, **iPhone** backup indicators, and even a live **Roblox** account valuation. To enhance the accuracy and value of this data, the malware exploits stolen session access to scour the victim's **Gmail** account for receipts from services like **Amazon**, **DoorDash**, **Uber Eats**, **Grubhub**, **Walmart**, and **PayPal**, using these to corroborate home addresses and spending habits. ### AI-Driven Victim Prioritization Once collected, the aggregated data bundle is serialized and fed into an AI system for analysis. This process leverages **OpenRouter**, a unified LLM API gateway, with requests defaulting to **Anthropic's Claude Opus 4.6**. If **Claude** refuses the request, the system resiliently falls back to **Moonshot AI's Kimi K2** and then **Google's Gemini 2.5 Flash**. **Talos** highlighted **VOLTA's** multi-model architecture: "VOLTA uses a multi-model architecture to fall back resiliently if it hits a model's refusal, checking each response for phrases like 'I can't' or 'I'm not able,' before passing on to a chain of up to three models." The malware's 240-word system prompt instructs the LLM to act as a "forensic intelligence analyst." This analyst is tasked with drafting a target assessment based *solely* on the provided evidence bundle, explicitly forbidding the model from using its training knowledge to supply facts about the subject. Key outputs from this AI analysis include a "WHAT THEY'RE WORTH" section, ranking valuables by monetization potential (low/medium/high tier), and a "THEIR OPSEC" section, assessing the victim's security posture and password hygiene. "The model is the operator's triage analyst; it turns a raw credential log into a prioritized and citation-backed dossier of the victim and their monetizable assets," **Talos** explained. ### Multi-Stage Data Exfiltration Data exfiltration occurs via **Telegram** in three distinct phases: 1. Immediately after the stealer runs, sending browser passwords, cookies, payment cards, cryptocurrency wallets, **Discord**, **Steam**, and **Roblox** tokens, Wi-Fi passwords, and screenshots. 2. Following the AI call, including the captured data and the results of the LLM analysis. 3. After the live enrichment step, which incorporates **Gmail** receipts and order histories. **VOLTA** also features a template with six configurable slots for build-time customization, including the **Telegram** channel for exfiltration and API keys. This suggests the malware author may be selling a builder kit or operating it as a bespoke service, compiling custom binaries for each order.  "If VOLTA is indeed a commoditized malware, then the AI dossier capability isn't something the author invented for their own operations," **Talos** stated. "It is a feature which enables a customer of any level of sophistication to point it at victims and start receiving ranked target assessments in their Telegram inbox." ### The Broader Trend of AI in Cyberattacks This development underscores a growing trend of adversarial adoption of generative AI (GenAI). Threat actors are increasingly integrating GenAI into their workflows to develop custom tools, identify valuable data, and enhance the success rate of their attacks. **Zscaler ThreatLabz** commented on this, noting: "GenAI technologies are also well suited to examine massive amounts of data for the purposes of classification or content searching. Once an attacker has access to the victim’s system, GenAI can therefore be used to search all materials to which a victim's system has access, seeking the most valuable or sensitive data for theft or encryption during the next phase of the attack." This method allows ransomware operators to pinpoint and acquire the most sensitive data, maximizing the likelihood of ransom payments. ### NIK C2: Another AI-Enabled Threat Last week, **Talos** also detailed **NIK C2**, an AI-enabled framework offering remote access, credential theft, keylogging, defense evasion, surveillance, and ransomware capabilities. **NIK C2** integrates an **OpenAI** client to send stolen data for AI analysis within the compromised session. This Go-based Windows remote access tool provides persistent control over compromised hosts, supporting over 52 commands from a command-and-control (C2) server. A notable feature is its **OpenAI** integration, which can abuse a victim's own **OpenAI** API key to fund the AI-driven analysis. If an API key isn't explicitly supplied by the attacker, the plant scans the victim's environment variables for "OPENAI_API_KEY" and uses it. This practice, known as LLMjacking, allows attackers to leverage another party's AI model, making the legitimate owner bear the computing costs and obscuring the attacker's identity. "If the victim has configured an OpenAI account on their machine, that account's API key will be used to fund the actor's AI inference," **Talos** warned. "NIK C2 reshapes adversary tradecraft by collapsing collection and exploitation into a single continuous session, effectively erasing the offline processing delays that historically granted defenders critical detection windows," **Talos** concluded. These sophisticated AI-driven tactics demand heightened vigilance and advanced defensive strategies from the cybersecurity community.