Malware uses LLMs to rank victims by how much they're worth

Saturday, October 31, 2026 08:00

• VOLTA is a mature stealer implant alongside an optional AI-integration layer. After stealing credentials, sessions, and cryptocurrency wallets, VOLTA sends a organized evidence bundle to an LLM to position the victim’s resources by monetization value.
• If a example refuses to execute the mark assessment, VOLTA fails complete to a three-model chain, implying a straightforward engineering attempt and anticipation of hitting example guardrails. 
• This stealer is self-deleting, and there is no attacker-owned infrastructure. The novelty in the AI-integration is that it is not supporting a traditional malware capability but operating as an intellect expert on behalf of the operator. 

There is a new idea in the cybercrime earth — mold a broad net, pilfer everything, and let AI kind out the specifics. A new malware self-named “VOLTA-AUTODOXER,” surfaced through the CAIRN investigation effort, is a self-deleting Windows infostealer that augments its post-theft workflow alongside an LLM-powered victim-profiling stage.

After harvesting the norm stealer haul, including browser passwords; cookies; fee cards; cryptocurrency wallet kernel phrases; Discord, Telegram, Steam, and Roblox sessions; Microsoft OAuth tokens; SSH and cloud-provider keys; Wi-Fi profiles; and screenshots, VOLTA builds a organized evidence bundle concerning the individual rearward the machine. This bundle contains geolocation and residence address, director and director domain, inferred commencement year, cardstock last-4 digits, Slack workspaces, document authorship metadata, iPhone backup indicators, and a live Roblox document valuation. It afterward uses taken meeting admission to enrich that bundle further: searching the victim's Gmail for acquisition receipts to confirm a residence address, pulling UberEats and DoorDash command history, and enriching Discord accounts via the taken token.

The bundle is serialized and POSTed to OpenRouter under a scheme immediate that instructs the example to act as "a forensic intellect expert example a mark assessment." The required output includes a division titled “WHAT THEY'RE WORTH” alongside valuables classified by monetization alongside a low/medium/high tier, and a division on “THEIR OPSEC” containing an appraisal of the victim's safety posture and password hygiene. The example is the operator's triage analyst; it turns a raw credential log into a prioritized and citation-backed dossier of the casualty and their monetizable assets.

The AI configuration: “Go deep, not wide... Be blunt, specific, accurate.” 

VOLTA uses a multi-model architecture to autumn rear resiliently if it hits a model’s refusal, checking all reply for phrases akin “I can’t” or “I’m not able,” before passing on to a sequence of up to three models.

Model temperate controls the randomness and predictability of LLM-generated text. The low hardcoded heat value (0.2) suggests the attacker is emphasizing accuracy and consistency. The malware additionally employs a sequence of cascading models in the event of failures. This multi-model architecture suggests that the controller is anticipating encountering refusals, alongside the fallback pivoting to a few, smaller, and additional permissive models.

The scheme immediate immediately reveals the operator’s use case: Produce a concise analytical appraisal of the victim’s suitability as a target, backed by an auditable trail of citations.

The immediate additionally asks for item out of the ordinary, but that agents excel at: making connections between sparse and disparate data. For example, encouraging connections akin the victim’s names, addresses, or universities to patterns in their passwords. In this case, the example isn't being asked to discover data; rather, it’s being asked to extend the data and create helpful inferences that broaden the value of what has been scraped.

Stealer capabilities 

VOLTA’s capabilities are immediately aligned alongside the records produced in contemporary digital life, and fine beyond essential credential theft. In a sole pass, the stealer can extract browser credentials, cookies, and saved fee cards throughout Chromium-based browsers. It hits Firefox and Thunderbird, Outlook profiles, Slack workspaces, Steam and Battle.net sessions, Discord tokens, Telegram sessions, Roblox accounts, and SSH and cloud-provider config records (gcloud, OpenVPN, Hide.me). VOLTA doesn’t fair copy cryptocurrency wallets, but runs an embedded cracker against the vault inline, penning the outcome to a WALLET VAULTS CRACKED row in its own report.  

Every saved Wi-Fi password on the device is pulled via a straightforward call to WlanGetProfileList and dumped to plaintext.

What really sets VOLTA distinct from commodity stealers is what comes following the first collection. There is a live enrichment continue that leverages taken sessions to profound dive the victim’s data. For example, using the victim's own authenticated Gmail tokens, it queries their inbox for receipts from Amazon, DoorDash, Uber Eats, Grubhub, Walmart, and PayPal, afterward assessment those receipts to established a confirmed residence address. Stolen Uber Eats and DoorDash sessions are called immediately for command history. The Discord token is re-used to enrich the victim's document profile. Roblox inventory is appreciated via a live “Rolimons[.]com” lookup.

After the gathering and enrichment stages are complete, the data is assembled and organized locally in a document called “Identity_Report.txt”. This document is serialized into a bundle as JSON and POSTed to OpenRouter alongside alongside the scheme prompt. The returned payload is received as “AI_Analysis.txt”. Both records exfil together in “volta_dox.zip”.

The conclusion round-trip happens entirely through the victim's own network connection, on their own machine, before the outcome is exfiltrated. 

Deployment model 

Instead of hardcoding its Telegram channel, API keys, and settings immediately into the code (which is apt what person penning a tool for individual use would do) VOLTA uses a template alongside six swappable slots, filled in at build time.  

This suggests it’s a builder model: The author sells access, the client provides their values, and the builder compiles a tradition binary for all order. The credentials in those slots are encrypted, but not in a meaningful way. The key is SHA-256 of two constants sitting in the identical binary, but it's presumably there since customers anticipate their API keys not to be sitting in plaintext, not since it provides genuine protection. Anyone who reverses the document can regain them.  

If VOLTA is certainly a commoditized malware, afterward the AI dossier capability isn't item the author invented for their own operations. It is a characteristic which enables a client of any flat of sophistication to item it at victims and commencement receiving classified mark assessments in their Telegram inbox.  

All the observed endpoints for this malware are third-party controlled. There is no actor-owned infrastructure.

Countermeasures 

Much of the token-stealing capability of VOLTA relies on techniques to bypass App-Bound Encryption (ABE) in Chrome. An use authority guideline that restricts which processes may initiate Chrome alongside --headless would obstacle the ABE decryption way entirely. At minimum, you can additionally vigilant whenever Chrome is launched alongside --headless from any procedure not in a known developer. Best practices for storing credentials in hardware backed stores additionally apply. 

In the event of compromise, the chief executable may have deleted itself, but users should attempt to maintain the “APPDATA%\volta_debug.txt” file. This is a timestamped crop log and contains a document of exactly what was collected and whether the LLM call succeeded. Revoke any possibly compromised tokens and alter passwords. For cryptocurrency wallets, if the kernel expression was in plaintext on disk, treat the wallet as completely compromised and migrate finances to a new wallet alongside a kernel generated on an air-gapped device. 

Conclusions 

VOLTA is a mature stealer alongside a late-stage AI tier added for one purpose: relieving the controller of analytical work. The axis of AI betterment current is not speed or measure but attempt displacement; the identical logic apparent in CLOSEDQUORUM, although applied to a distinct phase. These implants indicate the identical underlying logic: Identify the stage that requires the most sustained individual notice and substitute it alongside a example call. The inquiry for defenders is not whether this form volition appear elsewhere, but which phase it touches next.

The second trend to observe is the anticipation of hitting example refusals, and the approaches to construction in resiliency. In this case, it includes a fallback sequence from Claude to Kimi to Gemini and the capability to run without the LLM phase altogether if no key is present. As AI integration in malware matures, refusal resilience may rotate into a norm scheme requirement. 

The open inquiry underlying the two trends is whether attempt relocation in crimeware changes the landscape in a meaningful way: lowering the operational floor, raising the ceiling on scale, or both. Cybercrime operations have continually been opportunistic, and the sophistication bar has continually been lesser than assumed, but there is a difference between a broad net and a broad net that sorts itself.  

Sample details