WorkOS says attacker used ARTEX AI to breach internal sales app

What it is

WorkOS disclosed that on 30 October 2026 it found unauthorized access to an internal application used by its sales team. The company says the attacker used an AI offensive security framework to automate reconnaissance and exploitation attempts. Founder Michael Grinich identified the framework as ARTEX AI. WorkOS provides SSO, enterprise authentication and user management APIs, and its site lists at least 3,000 customers, including OpenAI, Anthropic, Cursor, Perplexity, Snowflake, Plaid and Netlify.

Preconditions / Affected

WorkOS lists four categories of exposed data:

• Business contacts from the CRM: names, email addresses, job titles and company affiliations.
• Customer account histories: product usage and connection or directory health.
• Customer communications: summaries and excerpts from sales calls, support discussions and customer Slack conversations.
• Account and relationship information about sales and customer relationships.

WorkOS says its investigation found no access to WorkOS app data or production services. A customer's exposure is the contact and relationship data held in the sales system, not the authentication data WorkOS processes for them.

What to do

• If you are a WorkOS customer, expect direct notification from WorkOS. Until then, treat inbound messages that cite your WorkOS usage, call content or contacts as possibly crafted from the stolen data. WorkOS warns of phishing and social engineering built on the exposed information.
• Search web logs for the user agents artex-enrich/1.0 and norma/0.4, which WorkOS found in the attacker's requests.
• Review access controls on internal sales and CRM tools, especially anything reachable from outside. WorkOS responded by disabling external access to the affected application and auditing access controls on its internal tools.

Caveats

WorkOS has engaged an independent forensics and incident response firm and has not yet determined the full impact. Grinich says the incident is contained and promises more detail after the investigation. The page does not say how the attacker got in.

The article describes ARTEX AI as released earlier in 2026 by Chinese developers as open source. It went private after attacks on South Korean banks, though several forks remain available. CrowdStrike reported that the group Vapor Panda used it against 200 organizations, and DIVD said it powered the attack it suffered in September. The article calls ARTEX AI the "likely" driver, and the attribution rests on the user agents and Grinich's statement. The page names no actor behind the WorkOS intrusion.