ARTEX campaigns from the inside:what 10 exposed consoles showed Check Point.
On 2026-11-05 Check Point Research published an analysis of ARTEX consoles whose full database was left open. The databases held operator prompts, model choices and agent actions, so the researchers could follow three criminal campaigns from the attacker's side.
- When
- Signing campaign 2026-10-14 to 10-29; report 2026-11-05
- Reported by
- Check Point Research; covered by eSecurity Planet (2026-11-09)
- Scope
- 10 exposed consoles; iOS signing shops, game account servers, 292 crypto domains
- ARTEX role
- The operators' main tool, with agents doing most of the work
What happened
Check Point found 2,557 systems that host or once hosted ARTEX panels, 1,608 of them live at the time. Ten were misconfigured so that their whole database was exposed. Those ten are the basis for the campaign analysis. Check Point describes the goals as opportunistic, financially motivated cybercrime and does not name a threat actor.
Campaign 1 targeted gray-market iOS app-signing shops. Two consoles ran for about two weeks, from October 14 to 29, used about 36 billion tokens and got more than 300 signing certificates with passwords from 10 shops. Three passed Check Point's checks for being usable.
Campaign 2 tried to drain crypto. Agents worked through 292 crypto payment domains and found exposed databases and personal data, but no money moved. In one case an unauthenticated web API leaked cloud storage keys with write access to a bucket serving iOS apps and landing pages. Check Point says the write was confirmed, the bucket owners later restored the original files, and no funds moved.
Campaign 3 went after servers behind gray-market game account tools. Agents recovered more than 190,000 Honor of Kings credential sets, including a plain text credential file left in a web root. A sample of about 3.3% was still valid, which Check Point projects to roughly 6,000 accounts. eSecurity Planet notes that figure is an extrapolation.
How ARTEX was involved
Operators mostly gave a goal and asked for progress, and let the agents do the rest. eSecurity Planet sums it up: the AI agents handled much of the work with little further human involvement.
Chinese models were the majority in the ten instances, with DeepSeek Flash models the most common. In one instance a guardrail fired on a US model and the operator switched to a more permissive Chinese model. eSecurity Planet adds that these observations do not establish a general safety comparison.
Some agent reports of success were wrong. Agents claimed working cash-out flaws based on API responses that said success, but the attacker's wallet shows no transactions. Check Point's words: not one cent moved to the attackers' pockets.
The weaknesses the agents found were common ones: admin passwords hardcoded in front-end code, web APIs without authentication, weak admin passwords and exposed files. Check Point quotes the lesson that it is not the complexity of attacks but the breadth and perseverance with which models test every endpoint.
Confirmed and alleged
Reported as observed
- 1,608 live ARTEX panels at the time of research, and 10 consoles with fully exposed databases (Check Point).
- More than 300 signing certificates with passwords taken; three passed usability checks (Check Point).
- No funds reached the attacker's wallet in the crypto campaign, and the S3 bucket owners restored the original manifest (Check Point).
Alleged, assessed or unclear
- The figure of about 6,000 usable game accounts is a projection from a 3.3% sample, not a count.
- Check Point groups some infrastructure by version tags, but warns that shared tags suggest similar builds and are not unique identifiers.
- Who the operators are. Check Point names no group and suggests one crypto operator may be a casual side earner rather than an established criminal.
Takeaways for defenders
Drawn from what the sources themselves recommend or observed. Not a complete checklist.
- Check Point's own advice: systems in the exposed perimeter should be tested and hardened accordingly.
- eSecurity Planet adds: confirm sensitive endpoints enforce authentication and authorization, remove exposed credentials and rotate them, restrict who can change downloads, manifests and landing pages, and alert on unexpected changes.
- Closing an exposed API does not cancel credentials already taken. Revoke tokens and sessions as well as resetting passwords (eSecurity Planet).
- Teams that run their own security agents should define authorized targets, limit credentials, keep action logs and require approval for disruptive steps (eSecurity Planet). This site's guide on authorized AI pentesting covers the same ground.
Sources
Every claim above comes from these pages or from this site's news data. Open them for the full reports.
- Hacking on Autopilot: Inside ARTEX Cybercrime Campaigns
Check Point Research, 2026-11-05
- ARTEX Cyberattacks: AI Agents Steal Credentials and Signing Certificates
eSecurity Planet, 2026-11-09
- Threat Overview of ARTEX Autonomous Penetration Testing Framework (FLASH-20261113-001)
FBI (IC3 FLASH, TLP:CLEAR), 2026-11-13
Follow the ARTEX story.
New forks, releases and reports, dated and sourced.
