Cisco Talos finds VOLTA-AUTODOXER infostealer that uses Claude, Kimi and Gemini via OpenRouter to profile and rank victims

Linha do tempo

1. securityaffairs.com ↗ · pub. 02/11/2026 15:34 · coletado 02/11 17:00

   VOLTA-AUTODOXER infostealer uses LLMs via OpenRouter to profile victims and rank their monetizable assets

   ler a análise desta fonte ocultar

   Cisco Talos analyzed VOLTA (self-named VOLTA-AUTODOXER), a self-deleting Windows infostealer that surfaced through the CAIRN research effort. After the usual data theft, it sends a structured evidence bundle about the victim to an LLM through OpenRouter. The model, prompted to act as a forensic intelligence analyst, returns a ranked dossier with sections titled "WHAT THEY'RE WORTH" and "THEIR OPSEC". Exfiltration goes through a Telegram channel. Talos assesses that the AI layer displaces analyst effort rather than adding speed or scale, and that the malware may be offered with a builder service.

   Technical Details

   • Theft scope: browser passwords, cookies and payment cards from Chromium browsers, plus Firefox, Thunderbird and Outlook data. It also takes Slack workspace data, Discord, Telegram, Steam, Battle.net and Roblox sessions, Microsoft OAuth tokens, SSH keys, cloud credentials, screenshots and Wi-Fi passwords. It tries to crack crypto wallet passwords locally. It bypasses Chrome App-Bound Encryption, which Talos ties to Chrome launched with --headless.
   • Profiling: it builds a profile of the victim covering location and home address, employer, estimated birth year, last four card digits and iPhone backup signs, and checks Roblox account value. It uses the victim's logged-in sessions to search Gmail for receipts (Amazon, DoorDash, Uber Eats, Grubhub, Walmart, PayPal) and to pull order histories.
   • LLM stage: requests are hand-built JSON with temperature 0.2. The default model is Anthropic Claude Opus 4.6. If it refuses (detected by phrases such as "I can't"), VOLTA falls back to Moonshot AI Kimi K2 and then Google Gemini 2.5 Flash. Without an API key the stealer still runs. The analysis runs on the victim host over the victim's own connection, with no attacker-controlled server.
   • Artifacts: Identity_Report.txt, AI_Analysis.txt, volta_log.zip, volta_dox.zip, and volta_doxnet.zip (when extra network data is found). %APPDATA%\volta_debug.txt is a timestamped log of collection and AI call results.
   • Builder: six configurable fields are encrypted with a weak scheme whose key is embedded in the binary, so credentials can be recovered by analysis.

   Remediation

   • Use application control to restrict which processes can launch Chrome with --headless, or at minimum alert on it outside a known developer set.
   • Store credentials in hardware-backed stores.
   • If compromised, preserve %APPDATA%\volta_debug.txt, revoke exposed tokens and change passwords.
   • Treat any wallet whose seed phrase sat in plaintext as fully compromised and move funds to a new wallet with a seed generated on an air-gapped device.
2. helpnetsecurity.com ↗ · pub. 02/11/2026 11:47 · coletado 02/11 12:01

   VOLTA-AUTODOXER infostealer uses LLMs via OpenRouter to rank stolen victim data by monetization value

   ler a análise desta fonte ocultar

   Cisco Talos disclosed a new Windows infostealer, VOLTA-AUTODOXER (VOLTA), found through its CAIRN research effort tracking AI-enabled malware. After a conventional theft phase, it sends the collected data to an LLM through the OpenRouter API. The model is asked to act as a forensic analyst, rank the victim's assets by how easily they can be monetized, and assess the victim's password hygiene. The malware deletes itself after running and uses only third-party endpoints, so there is no attacker-owned infrastructure to take down. Talos sees the same pattern, replacing the most human-intensive step with a model call, in the CLOSEDQUORUM implant.

   Technical Details

   • Collection: browser passwords, cookies and saved cards; cryptocurrency wallet seed phrases; Microsoft OAuth tokens; SSH and cloud provider keys; screenshots; Discord, Telegram, Steam and Roblox sessions; plaintext passwords of all saved Wi-Fi networks. It also includes a built-in cracker for cryptocurrency wallet vaults.
   • Enrichment: builds an evidence bundle with geolocation and home address, employer, inferred birth year, card last-4 digits, Slack workspaces, document metadata, iPhone backup indicators and a Roblox account valuation. Stolen sessions are used to search Gmail for purchase receipts, pull UberEats and DoorDash order history, and enrich Discord accounts via the stolen token.
   • AI analysis: data is assembled locally in Identity_Report.txt, serialized as JSON and POSTed to OpenRouter with a system prompt. The reply is saved as AI_Analysis.txt, and both files are exfiltrated in volta_dox.zip. Assets are tagged low, medium or high, and the model looks for links between personal details and chosen passwords.
   • Model fallback: the default model is Anthropic Claude Opus 4.6. Replies are scanned for refusal wording, and on refusal the job falls to Moonshot AI Kimi K2 and then Google Gemini 2.5 Flash. Talos reads this as the operator expecting refusals and falling back to more permissive models.
   • Builder model: configuration (Telegram channel, API keys) sits in six template slots filled at build time, with weakly encrypted credentials. This suggests the author sells a builder and compiles a custom binary per customer, delivering target assessments to the buyer's Telegram inbox.

   Remediation

   The article gives no specific mitigation. Defenders should expect AI-assisted triage of stolen data to spread to other phases of attacks.

3. thehackernews.com ↗ · pub. 02/11/2026 02:51 · coletado 02/11 04:00

   VOLTA-AUTODOXER infostealer uses Claude, Kimi and Gemini via OpenRouter to rank victims by monetization value

   ler a análise desta fonte ocultar

   Cisco Talos disclosed VOLTA-AUTODOXER (VOLTA), a Windows infostealer that uses an LLM to turn stolen data into a ranked dossier of each victim's monetizable assets. It harvests browser passwords and cookies, Outlook and Slack data, payment cards, crypto wallet seed phrases, Discord, Telegram, Steam and Roblox sessions, Microsoft OAuth tokens, SSH and cloud keys, Wi-Fi profiles and screenshots. The evidence bundle is sent through OpenRouter to Anthropic Claude Opus 4.6, with fallback to Moonshot AI Kimi K2 and Google Gemini 2.5 Flash. The output ranks valuables by monetization tier and assesses the victim's OPSEC. The article also covers NIK C2, an AI-enabled Go-based RAT that abuses the victim's own OpenAI API key (LLMjacking).

   Technical Details

   • The bundle includes geolocation and home address, employer and domain, inferred birth year, card last-4 digits, Slack workspaces, document authorship metadata, iPhone backup indicators and a live Roblox account valuation.
   • Stolen session access is used to search the victim's Gmail for Amazon, DoorDash, Uber Eats, Grubhub, Walmart and PayPal receipts to corroborate the home address.
   • A 240-word system prompt casts the model as a "forensic intelligence analyst" and requires "WHAT THEY'RE WORTH" and "THEIR OPSEC" sections.
   • Refusals such as "I can't" or "I'm not able" trigger fallback to the next of up to three models.
   • Exfiltration runs over Telegram in three stages: right after the stealer runs, after the AI call (data plus analysis), and after live enrichment (Gmail receipts and order histories).
   • A six-slot build-time template holds the Telegram channel, API keys and other settings, suggesting a builder kit or bespoke per-order builds.
   • NIK C2 supports over 52 C2 commands, with remote access, credential theft, keylogging, defense evasion, surveillance and ransomware capabilities. It falls back to the victim's OPENAI_API_KEY environment variable when the actor supplies no key.
   • Zscaler ThreatLabz notes that GenAI can be used to search a compromised system for the most valuable data to steal or encrypt.

   Remediation

   The article gives no explicit guidance. Defenders should protect browser-stored credentials and session tokens, and avoid leaving API keys such as OPENAI_API_KEY in environment variables on endpoints.

4. gbhackers.com ↗ · pub. 02/11/2026 02:08 · coletado 02/11 03:00

   VOLTA infostealer uses Claude, Kimi and Gemini via OpenRouter to profile victims and rank stolen data

   ler a análise desta fonte ocultar

   Cisco Talos, through its CAIRN research effort, identified a self-deleting Windows infostealer named VOLTA (self-labeled "VOLTA-AUTODOXER") that sends stolen credentials, sessions and crypto data to commercial LLMs via OpenRouter to produce a prioritized victim assessment. The AI layer ranks assets into low, medium or high monetization tiers and assesses the victim's password hygiene and security posture. It does not drive the initial theft. Observed communications rely on third-party services rather than attacker-owned infrastructure, and commercial distribution is not confirmed.

   Technical Details

   • The default model is anthropic/claude-opus-4.6, with fallbacks moonshotai/kimi-k2 and google/gemini-2.5-flash. The malware checks responses for refusal phrases such as "I can't" and "I'm not able", then moves to the next model.
   • Requests use manually assembled JSON, a browser-like User-Agent and a temperature of 0.2. The inference request originates from the compromised host before the dossier is exfiltrated.
   • Collection covers browser passwords, cookies, payment cards, crypto seed phrases, Microsoft OAuth tokens, SSH keys, cloud configurations, Wi-Fi credentials and screenshots. It also covers Discord, Telegram, Steam, Roblox and Slack, and an embedded component attempts to crack wallet vaults.
   • Enrichment reuses stolen sessions: Gmail receipts to corroborate home addresses, Uber Eats and DoorDash order histories, Discord tokens, and Rolimons lookups to value Roblox inventory.
   • Outputs are Identity_Report.txt and AI_Analysis.txt, packaged into volta_dox.zip. volta_log.zip and volta_doxnet.zip support other stages.
   • Six build-time configuration slots suggest a builder model. Customer credentials are encrypted with a key derived from constants in the executable, so they are recoverable through reverse engineering.
   • The App-Bound Encryption bypass path launches Chrome with --headless.

   Remediation

   • Monitor for unexpected processes launching Chrome with --headless, and restrict unauthorized headless Chrome execution with application controls.
   • Preserve %APPDATA%\volta_debug.txt, a timestamped log that survives executable deletion and shows whether the AI analysis succeeded.
   • Revoke exposed sessions and tokens, reset passwords, and move assets out of wallets whose seed phrases were compromised.
5. wired.com ↗ · pub. 31/10/2026 09:00 · coletado 31/10 11:00 primeiro registro

   Cisco Talos discovers Volta-Autodoxer infostealer that uses commercial LLMs to automatically profile victims

   ler a análise desta fonte ocultar

   Cisco Talos researchers identified a new infostealer, internally named Volta-Autodoxer, that feeds stolen victim data into commercial LLMs to build a prioritized, citation-backed dossier on each victim. The malware was flagged using Talos' new CAIRN framework, which is designed to identify hacking tools with AI capabilities. Talos did not track whether Volta has been used in the wild. Clues suggest it may have been built to be sold to a range of attackers rather than for a single actor's campaigns. The AI layer is optional, so the stealer still works without it.

   Technical Details

   • Collection: browser-stored passwords, Microsoft OAuth tokens, cookies, payment cards and crypto wallet seed phrases, plus active Discord, Telegram, Steam and Roblox sessions. The targeting appears geared toward younger users with extensive digital lives.
   • Second-stage sweep: using first-round findings, it searches the victim's Gmail or another central account for receipts and order history from services such as UberEats and DoorDash, to infer location and habits.
   • AI analysis: the serialized dossier is sent through OpenRouter to Anthropic Claude Opus 4.6 by default, with fallbacks to Moonshot AI Kimi K2 and then Google Gemini 2.5 Flash, apparently to work around model refusals.
   • Prompt: a 240-word prompt casts the LLM as a "forensic intelligence analyst" producing an executive summary, the victim's value ("what they're worth") and an opsec assessment, and asks it to connect dots across fields (for example, Wi-Fi password versus last name). The prompt is worded to avoid overtly malicious language.
   • Context: infostealer-sourced credentials have driven major breaches, including the Snowflake-linked incidents (AT&T, Ticketmaster) and the Disney data theft. CrowdStrike separately reported four hacking groups using AI tools to compromise dozens of targets worldwide.

   Remediation

   The article gives no specific remediation guidance. Because the purpose is faster triage of high-value accounts and reuse of stolen tokens, the practical risk is token and credential reuse before expiry or password changes.