Potheeswaran 
ServiceNow has patched a critical remote code execution (RCE) vulnerability in its AI Platform that, under certain conditions, could have allowed unauthenticated attackers to bypass the platform’s sandboxing controls and execute arbitrary code.
Tracked as CVE-2026-6875, the flaw was disclosed on July 13, 2026, via ServiceNow’s security knowledge base (KB3137947), with fixes already rolled out across hosted instances and made available to self-hosted customers and partners.
While ServiceNow says it has no evidence of active exploitation, the vulnerability’s severity, its unauthenticated attack vector, and the platform’s widespread enterprise footprint make this a patch-now priority for security teams.
ServiceNow Patches Critical AI
At the core of CVE-2026-6875 is a sandbox escape, a class of vulnerability in which code intended to run in an isolated, restricted environment breaks free and interacts with the underlying host system.
In the context of ServiceNow’s AI Platform, which powers Now Assist and other generative AI features embedded across the ecosystem, sandboxing is meant to contain AI-driven code execution, plugin logic, or workflow automation so it can’t touch sensitive backend infrastructure.
An escape here is particularly dangerous because AI platforms often process untrusted inputs, user prompts, third-party integrations, and dynamically generated scripts by design.
If that execution boundary fails, an attacker doesn’t need valid credentials to potentially run code within the broader ServiceNow environment, turning an AI feature into an entry point for full platform compromise.
ServiceNow’s advisory specifically notes this could be exploited by an unauthenticated user “in certain circumstances.” That phrasing is doing some heavy lifting for defenders: it means the usual first line of defense, requiring valid login credentials, doesn’t apply here.
For a platform that centralizes IT service management, HR case data, security incident workflows, and now AI-driven automation for thousands of enterprises, an unauthenticated RCE path is about as high-stakes as it gets on the severity scale.
ServiceNow has issued fixes across multiple family releases:
- Brazil (EA and GA)
- Australia (Patch 2)
- Zurich (Patch 7b and Patch 9)
- Yokohama (Patch 12 Hot Fix 1b and Patch 13)
Hosted instances have already received the update automatically. Self-hosted customers and partners need to apply the corresponding patch manually; details are available in KB2930717 (hosted overview) and KB2930740 (self-hosted overview), both gated behind a Now Support login.
- Confirm your instance’s release/patch version against the fixed versions listed above.
- Prioritize self-hosted environments, since these require manual patch deployment rather than automatic rollout.
- Audit AI Platform and Now Assist usage logs for anomalous execution patterns, even though no exploitation has been confirmed.
- Treat this as a template for future risk: as generative AI features are embedded more deeply into enterprise SaaS platforms, sandbox integrity becomes a frontline security control, not a backend implementation detail.
ServiceNow’s swift patch timeline and proactive disclosure are a good sign, but the underlying lesson stands: AI platform sandboxes are now high-value attack surfaces, and enterprises should treat AI-adjacent CVEs with the same urgency as traditional RCE flaws.
About the Author
