FBI Warns Criminals Are Using Artex to Go From Scan to Breach in Under Three Hours
Security / analysis
FBI Warns Criminals Are Using Artex to Go From Scan to Breach in Under Three Hours
Microsoft's own post says the 2,250 ConnectWise tenants were targeted, not all breached; Barracuda's timeline shows what machine speed looks like.

The FBI has told U.S. organisations that criminals are running Artex, an open source AI penetration-testing framework, against financial firms, governments and law offices, and that the tool has taken attackers from first scan to compromise in under three hours. The bureau's flash advisory (number 261113) was published on November 13, according to its filename, and was reported on November 16 by TechCrunch.
The advisory is a PDF with compressed text that The Terminal could not extract, so the details below come from TechCrunch's account of it and from the two incident reports it points to: Microsoft's analysis of the ConnectWise compromise and Barracuda's account of its own breach.
What Artex is and what it needs
Artex first became publicly available in July 2026. Microsoft describes it as "an open-source penetration testing framework that can use any AI model", a multi-agent system that runs from reconnaissance through data exfiltration. Operators can point it at hosted models from Anthropic and OpenAI, at open source models, or at non-U.S. models, per TechCrunch.
An attacker needs a network to scan and one weak spot: a password that is exposed or stolen, or a service without multi-factor authentication. TechCrunch reports the tool is noisy, generating traffic that can trigger alerts, but that it moves faster than a human defender can respond. The developer has since made Artex closed source. The FBI says the earlier open source copies remain "widely accessible", and researchers have found hundreds of Artex servers exposed online without passwords, including the one used against South Korea.
Three intrusions, three different front doors
| Victim | How the attacker got in | Reported scale |
|---|---|---|
| ConnectWise (Cloud Backup and other cloud apps) | Stolen application credentials, then customer tenants | Sign-in attempts against more than 2,250 tenants |
| Barracuda | Cloud access key left with a third-party AI vendor | About 10 cloud accounts in roughly 44 hours |
| South Korean financial institutions | Missing multi-factor authentication, weak defenses | Personal data of at least 68,000 people |
Microsoft attributes the ConnectWise activity to Storm-0940, which it assesses as aligned with the Chinese state, and says it began as early as September 2026. Microsoft also said no Microsoft product or service was compromised or vulnerable. ConnectWise published its first advisory on October 16. Barracuda, by contrast, says its match to Artex reveals the attacker's methods, not its identity, because many groups use the tool.
The 2,250 figure is attempts, not victims
TechCrunch writes that attackers accessed Microsoft 365 environments of at least 2,250 ConnectWise customers. Microsoft's own post says something narrower: Storm-0940 used Artex to attempt sign-ins to more than 2,250 tenants, and "the majority of the targeting wasn't successful". Separately, a password-spray run from October 10 to October 14 against 327 accounts at eight organisations produced more than 151,000 failed logons, peaking at 2,900 failures a minute on October 11, with no successful authentication observed.
Both can be true: some tenants were breached and most were only probed. The difference matters to a customer deciding whether to rotate credentials or only to read logs. Neither source gives the count that succeeded, so the number of compromised customers is unknown.
- ConnectWise tenants targeted2250 count
- Accounts in October password spray327 count
- Barracuda cloud accounts touched10 count
Source: Microsoft Security Blog, October 21, 2026; Barracuda blog, November 11, 2026
What Barracuda's timeline shows
Barracuda's engineers dated the first use of the stolen key to November 1 at 04:33 UTC. Each secret the attacker read led to the next system: cloud accounts, an analytics system, a source-code token, then Kubernetes clusters. Within about four hours the attacker pushed short-lived branches to several repositories to trigger a build job that printed build secrets into logs, encoded to slip past log masking. Last activity was recorded at 23:31 on November 2, and Barracuda's Chief Information Security Officer, Arve Kjoelen, wrote of the unretired key: "That is on us."

One system the attacker reached stores the credentials Barracuda XDR uses to collect customer data. Barracuda found no direct evidence they were read, but its logs were too thin to rule it out, so it rotated everything and asked customers with Managed XDR Cloud or Monitored Endpoint API integrations to revoke and recreate their credentials. TechCrunch says Barracuda gave no number for the "minority" of XDR customers it notified.
What to do
Microsoft's guidance for ConnectWise customers is to keep the affected application connections disabled until ConnectWise issues restoration instructions, search sign-in logs by application ID rather than display name, and review activity from at least October 15, extending back to October 1 where records allow. Barracuda's advice is to retire standing credentials, so one key cannot unlock the next system. The FBI advisory's own recommendations were not available to The Terminal in readable form; the PDF is at ic3.gov.
Related reading: the Microsoft 365 outage traced to a July gateway defect and Tencent's LoopForge exposing a Docker socket show how little a stray credential or open interface has to do before it is useful to an attacker.
Sources
More in Security
- 01Java4Shell Scores 9.2 From Its Finder and 8.1 From OpenJDK, and Needs an AVX-512 CPUCalif says a 65-character Base64 string overwrites 62 bytes of Java heap and reaches code execution in Keycloak, Jenkins and Jira. A second security firm says the source shows only a crash.
- 02CISA Gave 9 of 16 November KEV Entries a Three-Day Deadline With Forensic Triage, and 7 Got 14 DaysThe catalog's own due-date and forensicTriage fields show two tiers under BOD 26-04. Sorting November's additions by tier shows which exploited bugs CISA treats as already-compromised-until-proven-otherwise.
- 03Project Zero Cuts Its Disclosure Deadline to 60 Days on Feb. 2, 2027, Then 30 in 2028Google's bug-hunting team is dropping the 30-day patch soak period and counting every deadline from the day it reports, citing AI-assisted exploitation.
- 04Google Finds exploit_agent Diffing Patches and Firing Exploits With No Operator in the LoopGTIG's November 16 tracker reports the first fully automated discovery-to-exploit pipeline it has seen, found on an exposed staging server, plus 22 zero-days in August.