What happened. WorkOS, which sells login and identity plumbing to software companies, found unauthorized access to an internal sales app on October 30. Its bulletin says the attacker got business contact details, account histories, product usage data, and summaries and excerpts from sales calls, support threads and customer Slack conversations. WorkOS says there’s no evidence production services or application data were touched. Founder Michael Grinich tied the attacker to ARTEX, an open-source AI penetration testing suite, based on user agents like “artex-enrich/1.0” seen in the traffic, per BleepingComputer.
The second one. Chainalysis, the blockchain intelligence firm, says an external actor used “offensive agentic techniques” to get into its systems and internal tools. It says user IDs, names, emails and organization role labels were accessed, and it found no sign that investigation content or session tokens were. Its November 4 update adds something worse: an “on-path attack” exposed a subset of customer passwords. Chainalysis has not named a tool or shared technical details.
The pattern, in plain words. Neither company lost its crown jewels, as far as they know. Both lost the back office. The sales app. The internal tool. The admin view nobody put behind the same controls as production. That’s not an accident. An AI agent that can scan, try and retry all night doesn’t get bored. It finds the forgotten app with the weak login faster than any human crew would.
Why sales data is worth stealing. A customer list is old news. A customer list with real call notes and Slack excerpts is a phishing kit. If an attacker can write “following up on Tuesday’s call about your SSO rollout,” and it’s true, your people will click. WorkOS itself flags that risk. For a company whose customers are security and identity teams, that context is gold.
Where I’m careful. Attribution here is thin. WorkOS points to ARTEX through user-agent strings, which anyone can fake or copy. Chainalysis named nothing. And “AI-assisted” can become a convenient label that makes a breach sound inevitable. Stolen credentials and an exposed app are old problems. The AI part changes the speed and the scale, not the root cause.
What changes for defenders. The old idea was to protect production hardest and accept some looseness in internal tools. That trade no longer holds. Automated attackers make every exposed login page cheap to test. So the internal CRM, the support console and the sales dashboard need the same basics as prod: single sign-on, phishing-resistant MFA, no public exposure unless there’s a reason, and alerts when one account pulls far more records than usual.
For buyers. If you’re a customer of either company, assume your contact data and maybe your conversations are in someone’s hands. Warn your team that the next phishing email may quote a real thread. If you use Chainalysis and got a password notice, reset it and anywhere you reused it. And ask every critical vendor one question: are your internal tools held to the same bar as your product?
Where I land. AI attackers aren’t breaking new doors. They’re walking through old ones faster. The soft middle of a company, the tools between the product and the people, is now the front line. Treat it that way.