New VOLTA infostealer weaponizes mainstream AI models to help hackers mine victims’ data
A newly identified malware strain dubbed VOLTA goes beyond stealing passwords and crypto wallets: it taps models like Claude, Gemini, and Kimi to rank stolen files by value and auto‑generate target dossiers. The tool shows how attackers are now using the same AI systems enterprises rely on to triage victims faster and sharpen follow‑on fraud.
A new piece of malware is giving attackers their own AI assistant, using popular language models to sort through stolen data and tell them which victims are worth the effort.
Cybersecurity researchers have detailed an infostealer dubbed VOLTA that combines familiar theft techniques with something more recent: calls to mainstream AI models, including Claude, Gemini, and Kimi. Once it has infected a machine, VOLTA lifts passwords, browser cookies and cryptocurrency wallet information. It then scans the victim’s Gmail receipts to verify identity and purchasing history, and sends an AI‑generated profile of the target to attackers over Telegram.
For ordinary users and small businesses, the change is less about new data being stolen and more about what criminals can do with it. Traditional infostealers often leave attackers with vast troves of unstructured information. Sifting through that pile to find the one victim with a six‑figure crypto wallet, a business email account with wire‑transfer authority, or a cloud admin password takes time and manual labor. VOLTA’s use of AI models is meant to automate that triage.
According to the published technical breakdown, the malware feeds stolen data to the models, which then rank victims by their potential value. High‑value profiles – such as users with verified financial activity, business roles that suggest access to funds, or logins that touch sensitive infrastructure – are flagged and summarized. Those summaries, sent via Telegram, give attackers a quick briefing on who is on the other end of the stolen credentials and how best to monetize them.
Operationally, that can speed up the window between infection and follow‑on attacks like account takeovers, business email compromise fraud, or targeted extortion. If AI tooling tells a criminal that a given victim is a corporate finance officer with access to payment systems, that account may be used to send convincing but fraudulent invoices. If another victim appears to hold significant cryptocurrency, that wallet becomes a priority for immediate draining before defenders can react.
The fact that VOLTA leans on widely available AI models also matters for defenders. The models involved are general‑purpose systems, not dark‑web‑only tools. Any service reachable over the internet can, in theory, be scripted into an attack pipeline. That raises questions for AI providers about how well their abuse‑detection systems can spot and block patterns of use associated with malware, especially when requests may be routed through proxy servers and mixed with legitimate traffic.
For enterprises, VOLTA is an illustration of a broader trend: attackers are starting to adopt the same AI capabilities companies are racing to deploy for customer support, coding assistance and document search. The asymmetry cuts both ways. Defenders can use AI to scan logs and detect anomalies faster, but criminals can exploit the same flexibility to analyze stolen material and adapt social‑engineering lures in near real time.
One clear takeaway is that the core defenses against an AI‑assisted infostealer look a lot like those for older strains: keeping software patched, hardening browser and password storage, using hardware keys and multi‑factor authentication, and segmenting high‑value accounts so that a single set of stolen credentials can’t unlock everything. What changes with tools like VOLTA is how quickly and selectively attackers can act on whatever they manage to grab.
Signals to watch now include whether VOLTA or copycat malware begins to appear in large‑scale phishing or drive‑by download campaigns, whether major AI providers announce new guardrails or detection measures specifically aimed at this kind of misuse, and how often incident responders start seeing AI‑written victim dossiers in the command‑and‑control chatter of compromised networks.
Sources
- OSINT