Cisco Talos researchers have identified a new infostealer malware that incorporates artificial intelligence to automatically profile victims and identify the most valuable targets and their data. The malware, which calls itself Volta-Autodoxer internally, was found using Talos's new CAIRN framework, a system designed to detect hacking tools that use AI capabilities. Talos did not determine whether Volta has been deployed in actual attacks, according to security researcher Ryan Fetterman.
Volta operates as a mature stealer that gathers a broad range of data, Fetterman said. It assembles the collected information into a serialized dossier and sends it to multiple LLM models, asking them to function as an intelligence analyst and produce a structured, citation-backed assessment of the victim. Fetterman described the process as a prioritization and analysis function layered on top of the malware's core data theft.
The malware routes its analysis requests through OpenRouter, a service that can direct prompts to multiple AI providers. Volta defaults to Anthropic's Claude Opus 4.6, but includes fallbacks to the Chinese open weight Moonshot AI Kimi K2 and then Google's Gemini 2.5 Flash in case the first model's guardrails block the activity. Building in fallbacks and relying on cost-effective, older-generation models are common techniques among attackers.
Volta's creators embedded a 240-word prompt, including bullet points and a numbered list, instructing the LLM to act as a forensic intelligence analyst drafting a target assessment from a structured evidence bundle. The prompt asks for an executive summary covering what the victim does, what they are worth, and their opsec, and directs the model to connect details across fields, such as checking whether a Wi-Fi password matches a last name. It also tells the model to go deep rather than wide.
Fetterman said the prompt is carefully written to avoid overtly malicious implications, but that the creators anticipated model refusals on a first request and planned a cascade toward more permissive models. The implant also remains functional without the AI calls, which he characterized as an extra capability above the malware's core function.
Some clues suggest Volta may have been built for sale to a range of attackers rather than for one actor's campaigns, Fetterman said. While the malware remains a traditional infostealer at its core, the AI expansion points toward more invasive activity ahead. Fetterman noted that identifying high-value targets faster could increase the chances of using stolen credentials or authentication tokens before they expire or are changed, and said more advanced capabilities are reaching lower-level attackers.
Security researchers are increasingly finding hackers using LLMs for coding and for analyzing stolen data. Crowdstrike researchers recently told WIRED they had observed four distinct hacking groups of varying sophistication, operating in Chinese, Spanish and English, Indonesian, and Russian, using AI tools to compromise dozens of targets worldwide, including government entities and businesses in health care, retail, and travel.
More cybersecurity news from TechManNews.
