Image: cdn.prod.website-files.com · rights & removal
ARTEX and Storm-0940: How an AI Attack Framework Became a Growing Threat to Cloud Tenants
Reporting by Todyl Threat ResearchRead the original at todyl.com
Executive Summary
Facts Only
* ARTEX moved from a GitHub project to an attacker staple in about three months.
* Since late August 2026, ARTEX has been used by state-aligned espionage groups, financially motivated crews, and unnamed operators against cloud tenants, internet-facing servers, and connecting software vendors.
* Storm-0940 is assessed by Microsoft as a China-state aligned actor.
* The activity involved compromising a third-party vendor's Microsoft 365 applications via ARTEX and attempting sign-ins against customer tenants using valid credentials.
* Exposed secrets from AI coding agent settings files, including database passwords and API tokens, are targeted by ARTEX agents.
* Initial access for Storm-0940 involved a password spray targeting numerous accounts, resulting in hundreds of thousands of failed attempts.
* The attack involved gaining control of vendor app identities to run Microsoft Graph reconnaissance, adding credentials to OAuth applications, reading secrets across Key Vaults, and escalating privileges in one tenant.
* Sign-ins were attempted against more than 2,250 tenants using ARTEX.
* The activity has expanded to target other IT and software vendors focusing on identity providers like Okta and Entra.
Full Take
The narrative illustrates a shift where publicly available tools, when weaponized by autonomous agents, bypass traditional security assumptions regarding access control and supply chain trust. The underlying mechanism moves beyond exploiting specific vulnerabilities to leveraging exposed, pre-existing credentials—a pattern that acknowledges the reality that credential leakage is more common than organizations realize. The rapid adaptation of ARTEX, demonstrated by its ability to correct errors (retrying requests based on API versions) and adapt to detection (rotating fingerprints), suggests a trajectory toward autonomous systems that operate with high velocity and self-correction, challenging conventional reactive defenses built around slower human response times. The threat extends significantly to the MSP channel because the compromises target the implicit trust inherent in vendor-customer relationships; an actor compromising a single vendor can effectively compromise thousands of end-customers via delegated permissions. This maturation implies that security efforts must pivot from simply patching known flaws to fundamentally redesigning identity and access management within multi-tenant environments, focusing on limiting what compromised identities can reach rather than just blocking initial access vectors. The focus on service principals over interactive user sign-ins highlights a threat landscape where machine identities, lacking typical MFA safeguards, become the primary targets for automated exploitation.
BRIDGE QUESTIONS: If autonomous agents operate at speeds exceeding human reaction time, how can security architectures be designed to verify intent across complex, rapidly changing interactions rather than just verifying static access permissions? What systemic changes are required in vendor-customer relationship governance to mitigate risk when a single point of failure (a compromised application) can yield access to numerous downstream entities? What long-term strategic shifts are necessary for MSPs to move beyond managing per-client controls toward securing the entire extended identity fabric across their entire portfolio?
From the original · Todyl Threat Research
ARTEX, an open-source pen-testing framework that uses AI agents to carry out attacks with little human direction, has moved from GitHub project to attacker staple in about three months.Read the full story at todyl.com
Sentinel unavailable
The automated check of the source article's wording did not complete, so no result is shown.
This looks only at the wording of the original source article, not at this page's AI-written sections. A small local AI model made this estimate. It has not been checked against known human and machine texts, so treat it as provisional. It cannot show who wrote an article.
