What happened
CVE-2026-103093 (CVSS 9.4, CWE-284): ServiceNow remediated a sandbox-escape issue in the ServiceNow AI Platform where an authenticated user with the admin role could bypass Rhino sandbox restrictions and execute arbitrary code on the underlying host OS, potentially enabling further unintended access. ServiceNow deployed an update to hosted instances and provided patches to partners/self-hosted customers; no known malicious exploitation. Fixed in Yokohama Patch 13 Hot Fix 4a, Zurich Patch 10 Hot Fix 3b / Patch 12 / Patch 13, or Australia Patch 5.
Why it matters
A sandbox escape in an enterprise AI PaaS escalates from the AI workflow sandbox to full host code execution, threatening the platform instance and any data/credentials the AI workloads touch. Admin prerequisite limits reach but the consequence is host compromise.
Attack vector
An authenticated admin runs JavaScript/code in the Rhino sandbox that escapes restrictions to execute arbitrary commands or code on the host OS.
Affected systems
ServiceNow AI Platform before Yokohama Patch 13 HF 4a / Zurich Patch 10 HF 3b / Zurich Patch 12 / Zurich Patch 13 / Australia Patch 5.
Mitigation
Apply the patched release for your Instance version per ServiceNow KB3228994; restrict admin access to the AI Platform sandbox.