1-minute read
What a defender needs to know
ServiceNow patched CVE-2026-6875, a critical unauthenticated remote code execution flaw in its AI Platform. Searchlight Cyber found that the public /assessment_thanks.do page passed sysparm_assessable_type into a GlideRecord query. A javascript: expression gained restricted script execution, and a script-include trick escaped that sandbox.
The researchers said the resulting access could read platform tables, create administrators and run shell commands on connected proxy servers. Those servers often sit inside customer networks, so an instance compromise could extend beyond ServiceNow.
ServiceNow deployed initial cloud mitigations within 24 hours of the April 1 report and later released patches for hosted, self-hosted and partner environments. Its advisory said the company was not aware of exploitation when the CVE was published. That statement should be preserved as a limit on the public evidence.
Verify the fixed release for every production, nonproduction and internet-facing instance. Hunt for requests to the affected page with JavaScript expressions in the query parameter, then review administrator changes, sensitive table access and modified scripts. Include connected proxy servers in the investigation, especially ServiceNow-initiated processes and outbound connections. This was a flaw in the platform’s script and integration machinery, not a prompt-injection attack.