Agent News Daily

FBI warns attackers are using the ARTEX agent framework to speed up intrusions

An FBI FLASH alert says ARTEX, an open-source framework that drives AI agents, lets attackers reach privileged access in under three hours in some cases.

The FBI has issued a FLASH alert about ARTEX, an open-source framework that connects large language models to AI agents that carry out the steps of a network intrusion. The Record reported the alert on 13 November. It said the FBI described the tool as able to automate intrusion phases "at machine speed", from reconnaissance to data exfiltration.

What ARTEX is

Cyber Security News describes ARTEX as an open-source security testing framework written in Go, not a malware family. An operator sets an objective, and the agents scan targets, run tools, review the results and retry failed steps with limited human input.

According to The Record, the tool appeared on GitHub in August under an account called "Autumn-27". Citing The Wall Street Journal, it named the developer as Chinese security engineer Li Puhua. The developer said on 8 October that the tool was being abused and denied responsibility. The original repository was later removed, but The Record says forks, archived copies and a commercial version are still available.

What the FBI says it has seen

According to the two outlets, the alert describes the following:

  • Speed: some intrusions went from first targeting to privileged access in under three hours. Cyber Security News notes that this describes some observed cases, not a fixed capability.
  • Sectors: The Record lists financial services, government, technology, defence, semiconductors, energy, healthcare, manufacturing, telecommunications and legal.
  • Entry points: attackers favour less-defended systems such as partner portals, mobile backends, legacy services and third-party integrations.
  • Techniques: Cyber Security News reports SQL injection, missing API authentication, exposed passwords and abuse of trusted cloud connections. After getting in, the agents added credentials to applications, deployed web shells and tried to steal more credentials.
  • Models: ARTEX can use any model. The Record names Claude, GPT, DeepSeek, GLM, Grok and Kimi, and reports the FBI found Chinese models most common in attacker-controlled setups.
  • Exposure: Censys found more than 1,600 ARTEX consoles exposed online. Cyber Security News cautions that an exposed console does not by itself show malicious use.

Earlier reports

The alert follows earlier warnings. Cyber Security News says Microsoft linked ARTEX to Storm-0940, a group it describes as China-state-aligned. The group attempted access to more than 2,250 Microsoft 365 tenants using compromised ConnectWise application credentials. The outlet stresses that this figure counts targeted tenants, not confirmed breaches.

Canada's Cyber Centre published a bulletin on the same activity, last modified on 26 October. It said the framework was used against hundreds of organisations and that the attacks are not yet fully autonomous: people still choose targets, set objectives and oversee operations. Its main concern is a reusable framework that could support attacks on an industrial scale.

Cyber Security News also connects ARTEX to the recent WorkOS incident. WorkOS has attributed that access to a researcher's testing agent, not to criminal use.

Advice for defenders

Cyber Security News reports that the FBI recommends restricting exposed administration tools, reducing application permissions and scanning repositories for secrets. The Canadian bulletin adds:

  • phishing-resistant multi-factor authentication for important accounts
  • regular reviews of access granted to third-party cloud applications
  • prompt patching of internet-facing systems
  • practice responding to incidents that unfold within hours

Caveats

We could not read the FBI alert directly, so the details here come from The Record and Cyber Security News. The list of FBI recommendations comes only from Cyber Security News. The Record's coverage focuses on the threat and does not set out specific mitigation steps.

Sources

Sources