AI‑Powered “Volta‑Autodoxer” Infostealer Targets High‑Value Victims
Researchers say a new infostealer called Volta‑Autodoxer uses large language models to profile and prioritize stolen data.
New AI‑Enhanced Infostealer Emerges
Researchers at Cisco Talos have identified a fresh strain of infostealer malware that incorporates large language model (LLM) functionality. Internally the code refers to itself as “Volta‑Autodoxer,” a name derived from its self‑identification as a “Volta” stealer with an “autodox” profiling routine.
Traditional Stealing Capabilities
Volta operates like classic infostealers, harvesting a wide array of data from compromised machines. It sweeps browsers for saved passwords, extracts Microsoft OAuth tokens, cookies, payment‑card numbers, crypto‑wallet seed phrases, and active sessions on platforms such as Discord, Telegram, Steam and Roblox. After this first pass, the malware conducts a deeper scan of central accounts—most notably Gmail—to collect receipts, order histories and other records from services including UberEats and DoorDash, thereby constructing a richer picture of the victim’s location and habits.
AI‑Driven Victim Profiling
Once the data collection phase completes, Volta feeds the compiled “serialized dossier” to an AI service called OpenRouter, which can forward prompts to multiple providers. By default the malware sends its request to Anthropic’s Claude Opus 4.6. Anticipating possible model guardrails, the code includes fallbacks to the Chinese open‑weight Moonshot AI Kimi K2 and, if needed, Google’s Gemini 2.5 Flash. This layered approach mirrors common attacker tactics of using cost‑effective, older‑generation models.
The embedded prompt—about 240 words long and formatted with bullet points and a numbered list—asks the LLM to act as a forensic intelligence analyst. It directs the model to produce an “executive summary” covering the victim’s activities, estimated “worth,” and an assessment of their operational security, even urging the AI to “connect dots across fields” such as matching a Wi‑Fi password to a last name. The prompt is crafted to avoid overtly malicious language while still extracting actionable insights.
Research Findings and Context
Talos researcher Ryan Fetterman described Volta as “pretty mature as a stealer” that “casts a pretty wide net for what it collects.” He noted that the malware can operate without AI, but the AI calls “are kind of above and beyond.” The team flagged Volta using their new CAIRN framework, designed to spot hacking tools that embed AI capabilities. Talos has not observed Volta in the wild yet.
The researchers also observed that Volta appears built for resale to a variety of attackers rather than a single campaign. This aligns with broader trends: other security firms, such as Crowdstrike, have reported multiple hacking groups across languages using AI tools for coding and data analysis, targeting entities in health care, retail, travel and government sectors.
Implications for Threat Landscape
According to Fetterman, faster identification of high‑value targets could increase the likelihood that attackers successfully reuse stolen credentials or tokens before they expire. He warned that “we’re seeing more advanced capabilities accessible to lower‑level sophistication attackers,” calling the development both “interesting and disturbing.”
While Volta’s core functionality remains a conventional infostealer, its AI‑enhanced profiling marks a notable evolution that could enable more invasive and efficient exploitation of compromised users.
