Cisco Talos describes a Windows information stealer that collects credentials and personal records, then optionally asks a language model to rank the victim’s assets. In Ryan Fetterman’s analysis, the program calls itself VOLTA-AUTODOXER. The model’s role is to sort information already stolen, not to perform the initial compromise.

Evidence status: vendor analysis of a malware sample. Talos describes capabilities in the sample and gives a first-seen date of October 11. The report does not establish how many victims were affected or whether every described stage succeeded in real attacks. Cisco sells security products and services.

What the sample collects

Talos says the stealer targets browser passwords and cookies, saved payment details, messaging sessions, cloud and SSH credentials, and other local records. It can also use stolen sessions to gather more information, including purchase receipts. The resulting dossier can be sent through a model service for an assessment of assets and security habits, then sent to the operator along with stolen data.

The model stage is optional. Talos reports that the sample can proceed without it and can try other models if one refuses. Its analysis says the sample uses third-party services rather than attacker-owned infrastructure. That makes blocking one obvious command server an incomplete defense.

Respond to a suspected compromise

The report concerns a Windows executable; it does not identify a universal delivery route or say ordinary browsing alone causes infection. A person who has run a suspicious program should treat browser sessions and credentials accessible to that device as potentially exposed.

  • Isolate the affected device and preserve evidence. Talos says the executable may delete itself but a timestamped debug log can record what was collected and whether the model call succeeded. Handle potentially sensitive logs carefully.
  • From a clean device, revoke exposed sessions and tokens and change affected passwords. Prioritize email, cloud accounts and developer credentials; review account recovery settings and sign-in activity. A password change alone may not revoke every session.
  • Review the scope of local secrets. Rotate accessible SSH and cloud credentials at their issuers, and check downstream access records. Coordinate organizational response before destroying evidence.
  • Consider application controls and monitoring. Talos recommends restricting or alerting on unexpected headless Chrome launches, which it says are involved in this sample’s browser-secret extraction. Legitimate development and automation can also use headless Chrome, so tune controls to the environment.

Caveat: The report analyzes one sample; it does not measure the prevalence or effectiveness of AI-assisted victim ranking across malware.