Skip to content

ServiceNow AI Platform CVE-2026-6875 exploitation

Summary

CVE-2026-6875 is a critical pre-authentication sandbox escape in the ServiceNow AI Platform that can permit arbitrary code execution. ServiceNow released security updates throughout June 2026, and Defused Cyber reported in-the-wild exploitation on 2026-07-21. The vulnerability is separate from the June 2026 hosted-instance table-query incident.

Public reporting attributes discovery to Searchlight Cyber and describes an unauthenticated route through the platform sandbox to operating-system code execution and complete compromise of the instance and connected proxy servers. Defused reported exploitation through HTTP POST requests to /assessment_thanks.do. Public actor attribution, payloads, victim count, source addresses, and hashes were not available at publication time.

Tags

Why this matters

  • Exploitation is unauthenticated and can cross an application sandbox into arbitrary host code execution.
  • ServiceNow is a high-value enterprise control plane with workflow, CMDB, ticket, identity, integration, and automation access.
  • The lack of public IOCs means defenders must prioritize version/advisory validation and behavior-based compromise assessment rather than waiting for a signature.

Publicly reported scope

  • Vulnerability: CVE-2026-6875, pre-authentication sandbox escape / arbitrary code execution.
  • Vendor advisory: ServiceNow KB3137947; updates were released throughout June for Brazil EA/GA, Australia Patch 2, Zurich Patch 7b/9, and Yokohama Patch 12 Hot Fix 1b/13.
  • Observed route: HTTP POST requests to /assessment_thanks.do; the reported sandbox-escape gadget reaches the same code-execution primitive through a different path than the public proof of concept.
  • Potential impact: complete compromise of the ServiceNow instance and connected proxy servers, according to Searchlight Cyber.
  • Exploitation status: Defused Cyber reported observed in-the-wild exploitation on 2026-07-21.
  • Unknowns: no public actor, malware family, victim count, source addresses, or payload hashes were established in the reviewed public reporting.

Detection and response

  1. Validate every affected ServiceNow AI Platform deployment against KB3137947 and apply the vendor-provided fixed release or update without delay.
  2. Restrict internet access to administrative and AI execution surfaces where business requirements permit; enforce identity-aware access and network allow-listing.
  3. Review reverse-proxy, WAF, platform, integration, and host telemetry for unauthenticated requests followed by sandbox errors, child-process creation, shell or interpreter execution, outbound connections, file writes, credential access, or new persistence.
  4. Scope ServiceNow integration credentials, OAuth applications, MID Servers, service accounts, secrets, and downstream automation reachable from the affected platform.
  5. Preserve request, platform, audit, process, and network evidence before remediation. On credible exploitation, isolate affected components, rotate exposed credentials, and coordinate forensic scoping with ServiceNow.
  6. Keep this incident distinct from the June table-query issue so that patch validation and compromise windows are not conflated.

Sources