Identity and authentication infrastructure provider WorkOS has confirmed that unauthorized attackers gained access to an internal sales application, exposing customer contact information and records of business conversations. The company says it detected the intrusion on October 30, 2026, and published a security bulletin disclosing the incident on November 2.

What sets this breach apart from a typical CRM compromise is the tooling behind it. WorkOS’s investigation identified an AI-assisted offensive security framework used to automate reconnaissance and exploitation against the company’s systems. Founder and CEO Michael Grinich went further than the official bulletin in a public statement, naming the tool as ARTEX — an AI-powered platform built to automate offensive security tasks with minimal human direction.

What Was Exposed

The compromised application gave attackers access to a range of customer relationship data, including:

• Names, email addresses, phone numbers, job titles, and company affiliations.
• Customer account histories and sales relationship records.
• Summaries and excerpts from sales calls and support discussions.
• Customer Slack conversations handled through the affected system.
• Operational metadata, including product usage patterns and the health of customer directory and connection configurations.

WorkOS is careful to frame these as categories of information the application made available, not a confirmed, record-by-record account of exactly what was taken from every customer. The company says it is still reviewing affected records to establish impact on individual organizations, and maintains that it currently has no evidence its core application data or production services — the infrastructure customers actually integrate with for authentication and identity — were accessed or affected. The investigation, as of the disclosure, points specifically to CRM and customer support systems rather than WorkOS’s production platform.

ARTEX: A Growing Pattern in Agentic Attacks

ARTEX isn’t an isolated curiosity. Research from Google’s Threat Intelligence Group and Mandiant describes it as a multi-agent offensive platform that converts broad, high-level attacker goals into specific technical commands — effectively letting an operator describe an objective and have autonomous agents handle the mechanics of getting there. According to that research, ARTEX’s agents are capable of testing permission boundaries, mapping internal networks, adapting automatically when a command fails, and progressing through multiple stages of an intrusion with limited ongoing human supervision.

Separate security research, including analysis from Wiz, has documented ARTEX-linked activity involving agents that download internal application files, enumerate undocumented API routes, and probe authentication controls — all tactics consistent with an automated reconnaissance phase. Other coverage has tied ARTEX-style tooling to campaigns targeting Microsoft 365 environments, where stolen application credentials were used to pivot into additional customer systems. It’s worth stressing that none of this broader research establishes attribution for the specific WorkOS incident — it simply describes the kind of tool now apparently being used against it, and indicates WorkOS is not an isolated target.

Why Exposed Conversations Matter More Than They Might Seem To

Stolen contact details are one thing, but the exposure of actual conversation content — sales call summaries, support tickets, Slack messages — raises a subtler risk. Attackers who can read how a company’s support and sales teams actually talk to customers gain the raw material to craft far more convincing follow-up phishing attempts, ones that reference real past conversations, real employee names, and real account details rather than generic pretexts.

WorkOS is explicitly warning its customers about exactly this risk, advising them to independently verify any unexpected request — even one that appears to reference a genuine prior conversation or cites details that seem to confirm it’s legitimate — before acting on it or sharing further sensitive information.

WorkOS’s Response So Far

Since detecting the intrusion, WorkOS says it has:

• Disabled external access to the affected sales application.
• Begun a broader review of access controls across its internal tooling.
• Retained an independent digital forensics and incident response firm to assist the investigation.
• Committed to contacting affected customers directly as findings are confirmed.

For organizations that use WorkOS for authentication or identity management, the practical takeaway is to treat any unusual or unexpected outreach referencing a WorkOS relationship with extra scrutiny for the time being, and to watch for a direct notification from the company if their account was among the affected records.