MediumBreach / campaignOngoing

SeverVault infostealer marketplace sells logs with AI profiles

A criminal marketplace sells infostealer logs with AI-written owner profiles; reset exposed credentials and revoke active sessions.

Published
Exploitation
Yes, Hudson Rock matched advertised logs to its own database, some from infections in the previous 24 hours, but it did not confirm completed purchases

What happened

Hudson Rock found a criminal marketplace, advertised on cybercrime forums, that sells data stolen from computers infected with infostealer malware. Each log comes with an AI-written profile of the machine’s owner, covering where the person appears to work and which of their accounts are worth buying, according to Help Net Security. Buyers can search these descriptions in plain language, for example “citrix access to healthcare provider”. Filters include country, city, ISP, stealer family, operating system, date and price.

On naming, the outlets differ. Help Net Security reports that Hudson Rock does not name the service. The headline of SOCRadar calls it SeverVault. The SOCRadar article body was not available in full to us, so we cite it only for that name.

Details reported by Help Net Security:

  • The stolen logs carry browser passwords, cookies and autofill data. Some include Citrix and VPN logins that could give a buyer a way into the victim’s workplace.
  • The operator also runs a separate catalog that sells individual logins, passwords and cookies by quantity.
  • The operator claims tens of millions of logs and more than a billion parsed accounts, with prices starting at $0.25 and a minimum cryptocurrency deposit of $100. Hudson Rock has not verified these inventory figures.
  • Example listings include a workstation at a Canadian ophthalmology clinic for $10 (11 saved passwords, 574 cookies) and a defense contractor’s household computer for $9, scored 100 out of 100. The listings carry tags and numerical scores, but no methodology for the scores was shown.
  • One seller claimed a listing from an AI researcher contained working session tokens for Anthropic’s Claude API. Hudson Rock could not establish what kind of tokens they were or whether they still worked.
  • Account searches returned results for uber.com and fbi.gov, which does not mean either organization was breached.

Hudson Rock matched several advertised logs to its own database, and some came from computers infected within the previous 24 hours. Alon Gal, Hudson Rock’s CTO, said it has not established the share of fresh versus recycled data and has not independently confirmed completed purchases. He said the AI descriptions still require validation but reduce the work of choosing targets.

Gal also confirmed that Hudson Rock is finding AI credentials in stolen logs and that infostealers target AI development tools on purpose. An API key can run up charges and reach files in its workspace, while a valid reusable browser session could expose conversations, uploads and projects, he said.

Am I affected?

The sources do not give a list of victims or a way to check exposure, so this is not confirmed for any specific organization. The following groups carry the most relevant risk, based on what is reported:

  • Anyone whose computer was infected with an infostealer, especially if browser-saved passwords, cookies, Citrix or VPN logins were present on it.
  • Organizations whose staff use personal or family computers for work logins, as in the listings described above.
  • Users who keep AI tool credentials (API keys, OAuth details, browser sessions) on machines that could be infected.

Gal noted that many services do not properly monitor infostealer infections or promptly reset exposed credentials and revoke sessions, so a detected infection does not mean the affected services have addressed it. Being listed in a marketplace search result does not by itself show that an organization was breached.

What to do now

  1. Treat any machine with a known or suspected infostealer infection as exposed, including every password, cookie and session stored in its browsers.
  2. Reset the credentials used on those machines and revoke active sessions, since Gal said unrevoked sessions and unreset credentials leave stolen logs usable.
  3. Review remote-access logins such as Citrix and VPN for accounts tied to affected users. The sources report these logins appear in some logs.
  4. If AI tool API keys or OAuth details were stored on an affected machine, treat them as exposed. The sources say a stolen API key can be used to run up charges and reach workspace files.
  5. Check whether your own services monitor infostealer infections and act on them. This is the gap Gal described.

Sources