New Infostealer Malware Uses AI to Profile Victims, Cisco Talos Finds

Cisco Talos researchers have identified a new infostealer malware strain called Volta-Autodoxer that uses commercial AI models to analyze stolen data and profile victims. The malware collects a wide range of credentials and session data, then feeds its findings into AI tools through OpenRouter, with Anthropic Claude Opus 4.6 as the default model. Talos did not track whether the malware has been used in the wild.

How Volta Collects Data

Volta sweeps for passwords stored in browsers, Microsoft OAuth tokens, cookies, payment cards, crypto wallet seed phrases, and active Discord, Telegram, Steam, and Roblox sessions. It then performs a deeper sweep through a target's Gmail or another central account, building a fuller picture of the victim, their location, and their habits using data such as receipts and order history from services including UberEats and DoorDash. The malware compiles its findings into a serialized dossier and sends them to LLM models, asking them to act like an intelligence analyst and produce a structured, citation-backed assessment.

AI Analysis and Model Fallbacks

Volta uses OpenRouter, a service that can send prompts to multiple AI providers. It defaults requests to Anthropic's Claude Opus 4.6. If model guardrails block the activity, it falls back to Moonshot AI Kimi K2 and, if that fails, Google's Gemini 2.5 Flash. The creators included a 240-word prompt instructing the chatbot to act as a forensic intelligence analyst and produce an executive summary, details of what the person does, what they are worth, and a summary of their opsec. The prompt also tells the chatbot to connect dots across fields, such as checking whether a Wi-Fi password matches a last name, and to go deep rather than wide.

Implications for Cybercrime

Talos flagged the malware using its CAIRN framework, built to identify hacking tools that incorporate AI capabilities. Ryan Fetterman, a security researcher at Cisco Talos, said Volta contains clues that it may have been built to sell to a range of attackers rather than for one specific actor's campaigns. The AI calls are optional and the malware can still work without them. Fetterman said the ability to identify high-value targets faster could help attackers use credentials or stolen tokens before they expire or are changed, and that more advanced capabilities are becoming accessible to attackers with lower sophistication.

Sources: Wired