WorkOS says attacker used the ARTEX AI framework to break into an internal sales app

Identity platform WorkOS has disclosed that an attacker gained access to an internal application used by its sales team. The company detected the intrusion on October 30 and described it in a security bulletin published on November 2, according to Cyber Security News. The bulletin referred to an AI-assisted offensive security framework that automated reconnaissance and exploitation attempts without naming it; chief executive Michael Grinich later identified it on X as ARTEX.

What was exposed

The affected tool held customer relationship management data: names, email addresses, phone numbers, job titles and company affiliations, along with account histories and sales relationship information. It also stored summaries and excerpts of sales calls, support discussions and customer Slack conversations, plus operational metadata on how customers use WorkOS products, such as connection and directory health. Cyber Security News notes that this list describes what the application contained, not a confirmed record-by-record account of what the attacker took.

WorkOS says there is no evidence that its application data or production services were accessed or affected. It has disabled external access to the sales app, is reviewing access controls across its internal tools, has hired an outside forensics and incident response firm, and plans to contact affected customers directly as findings are confirmed. Because the exposed material includes real conversation details, the company advises customers to verify unexpected requests even when they mention genuine colleagues or past support exchanges, since such details can make phishing more convincing.

What ARTEX is

According to SOCRadar, ARTEX is a multi-agent penetration-testing framework written in Go and driven by large language models. It was released as open source on GitHub in July 2026 by a developer based in China, who made it closed source on October 8, citing misuse; public forks remain, and a paid edition has appeared. Google Threat Intelligence Group and Mandiant describe agents that can probe permissions, map networks and retry failed commands with limited human input.

Open questions

The framework has turned up in several recent investigations, including CrowdStrike's report on a China-nexus group and Wiz research on another actor, but WorkOS has not attributed its incident to any group and those campaigns' techniques are not confirmed as part of this breach. The company has not said how the attacker got in or how many customers are affected, and its review of the exposed records is still under way.