ServiceNow AI Platform Sandbox Escape RCE Analysis

CVE-2026-6875 is a critical sandbox escape vulnerability in the ServiceNow AI Platform with a CVSS score of 9.5. This vulnerability allows an unauthenticated, network-adjacent attacker to bypass script execution restrictions and achieve full platform-level code execution. While ServiceNow has mitigated hosted instances, self-hosted and partner-managed deployments remain at high risk, especially following reports of active in-the-wild exploitation using alternative gadget chains.

Technical analysis reveals the flaw resides in the processing of the `sysparm_assessable_type` parameter at the `/assessment_thanks.do` endpoint. Attacker-controlled values can enter a JavaScript evaluation context (`GlideSystemSandbox`). By leveraging global mutable state and the `gs.include()` mechanism, attackers can transfer capabilities from a restricted sandbox to a more privileged script-loading context, eventually leading to full administrative control of the instance.

Successful exploitation grants access to sensitive platform tables, user credentials, and connected automation workflows. Critically, researchers have demonstrated that this access can extend to configured MID Servers, potentially allowing shell command execution within internal customer networks. Organizations should prioritize patching and perform immediate audit log reviews for any unauthenticated requests to the vulnerable sink.