If your defence plan assumes attackers need days to move from a forgotten login to admin rights, the FBI is telling you to revise it. A new advisory says an AI-driven attack framework called ARTEX has shortened some intrusions to under three hours, and attackers have already used it in the breach of ConnectWise’s cloud backup service. The weak points it finds best are the ones most teams stop watching: partner portals, legacy services and app integrations.

What happened

According to BleepingComputer, the FBI issued a FLASH advisory on Friday saying multiple threat actors have used ARTEX since at least August 2026. Targets span finance, government, technology, defence and healthcare. ARTEX is an open-source framework written in Go. It does not ship its own model. It connects to outside LLMs such as Claude, GPT, DeepSeek, GLM, Grok and Kimi, and turns an operator’s goals into tasks for specialised agents covering reconnaissance, exploitation, privilege escalation and data theft.

The developer, known as Autumn-27, posted it to GitHub in July and removed it on 8 October after abuse reports. The FBI says it persists through forks, archived copies, localised versions and a commercial edition, ARTEX PRO. Researchers had found more than 1,600 exposed ARTEX consoles by early November, and the true number is believed to be higher.

The advisory cites the October ConnectWise Cloud Backup breach, a service formerly called SkyKick Cloud Backup. Attackers compromised the service’s application identities and used the permissions customers had already granted. Huntress counted at least 162 compromised tenants. Microsoft’s investigation attributed the campaign to Storm-0940, a China-state-aligned actor, and said it used ARTEX to speed up reconnaissance, password spraying and abuse of cloud app permissions. Per that investigation, the attackers tried to authenticate against more than 2,250 downstream tenants after compromising the ConnectWise identities.

The FBI also points to attacks on South Korean banks that exposed data on about 67,000 people, a police case involving seven arrests and roughly 8.2 million stolen records, and a CrowdStrike report on Vapor Panda targeting more than 200 organisations in 14 countries.

Why it matters

The FBI says the framework lets operators with minimal technical expertise run complex, multi-target campaigns at once. This suggests the main change is in scale rather than technique: one operator can point agents at many victims in parallel, and organisations with unglamorous exposure may face attention they once avoided. That is our inference, not an FBI finding.

The ConnectWise case shows where this bites hardest. Trust granted to a vendor’s app identity works as a standing credential, so an attacker who compromises it can authenticate to every customer tenant that approved it. Pairing that with automated attempts across thousands of tenants can turn one vendor compromise into a wide campaign. Our earlier coverage of ScreenConnect abuse described a similar MSP-to-customer path.

Takedowns also look weak here. Removing the original repository did not stop distribution, and the framework relies on commercial models that defenders do not control.

What it means for you

Start with identity, not the perimeter. List every third-party application with consent in your Microsoft 365 or Google tenant, note the permissions each holds, and remove any you cannot justify. Ask backup, RMM and other vendors how they protect their own application credentials and how fast they would notify you of a compromise.

Next, inventory forgotten internet-facing systems such as partner portals, employee apps and legacy services, and retire or lock down what you find. Restrict access to cloud admin interfaces and add rate limits to sensitive APIs. Review authentication logs for bursts of automated requests, unexpected credential or privilege changes and large data transfers.

Because attacks can reach privileged access in under three hours, give alerts an owner who can act out of hours, with pre-approved steps to disable an app or reset credentials. Load the FBI indicators and the open-source detection pack from researcher MalwareMustDie into your monitoring, but treat IP addresses as perishable and expect some false positives. Report suspicious activity to the FBI or IC3. If you run an MSP, our piece on trusting one RMM console covers related concentration risk.

What to watch

  • Whether model providers tighten controls on how offensive agent frameworks use their APIs.
  • Growth in exposed ARTEX consoles beyond the 1,600 found by early November, and uptake of ARTEX PRO.
  • Further supplier breaches that abuse application identities across customer tenants.
  • Updates to the FBI indicators and the detection pack as infrastructure changes.

For more coverage, see our cybersecurity page.

Sources