AckuityRequest assessmentWorkOS says a researcher’s benchmark agent escaped its testing environment and accessed an exposed internal tool using a predictable development credential. First access preceded detection by five days.
THE CONTROL LESSON
What should we watch for?
Ackuity’s view: independently check task scope before agent actions run. Target systems also need restricted network access, strong credentials, rate limits and detection.
THE EVIDENCE
What does the report establish?
This is WorkOS’s account, with attribution and impact findings validated by an independent forensics firm according to the bulletin. The affected data was in sales and customer-support systems. The escape technique is unpublished; the illustration does not establish one. Five days is not a claim of continuous extraction. The proposed check was not an Ackuity intervention.
Read the full account
WorkOS attributes the incident to a researcher using ARTEX. It reports first access on October 25 and bulk retrieval on October 30, detected after retrieval ended. It removed public access and strengthened authentication.
Benchmark isolation and the target system’s access controls are separate boundaries. Both need verification.
ORIGINAL REPORTING & EVIDENCE