The ServiceNow RCE flaw, CVE-2026-6875, lets an unauthenticated attacker escape the platform’s script sandbox and run code on a self-hosted instance. Attackers began probing for it on July 17, 2026, just four days after disclosure. Therefore, teams should patch now, then hunt for new admin accounts and MID Server activity.
What Is CVE-2026-6875?
ServiceNow published advisory KB3137947 on July 13, 2026, and described a remote code execution flaw in the ServiceNow AI Platform. Specifically, an unauthenticated user can run code inside the platform under certain conditions. The flaw scores 9.5 Critical under CVSS v4.0 and maps to CWE-94, code injection. However, the vector also rates attack complexity as high, which explains why exploitation took some skill.
Researchers reported the bug to ServiceNow on April 1, 2026. Within about 24 hours, ServiceNow pushed protections to its hosted instances, and then it shipped patches to self-hosted customers and partners through June. In other words, the July 13 advisory marked public disclosure, not the first fix.
How Does the ServiceNow RCE Sandbox Escape Work?
ServiceNow lets customers run JavaScript inside sandboxed contexts. The attack starts at /assessment_thanks.do, an endpoint that handles survey and assessment responses and needs no login. First, the attacker sends a POST request with a sysparm_assessable_type value that begins with javascript:. Next, the platform evaluates that filter value as code. Then a gadget chain redefines a global function that a script include later calls, and because script include contexts sit outside the sandbox, the attacker’s code runs with platform privileges.
This is a classic case of attackers who exploit a public facing application without any credentials. Moreover, the public proof of concept uses one gadget route, while the attackers in the wild used another route to the same primitive. As a result, a single signature cannot cover the whole flaw.
Which ServiceNow Releases Fix the Flaw?
ServiceNow lists ServiceNow RCE fixes for four release families. Meanwhile, it treats any self-hosted or partner-hosted instance on an earlier release as vulnerable. Check your release against this table.
| Release family | Fixed in |
|---|---|
| Brazil | Brazil EA or Brazil GA |
| Australia | Australia Patch 2 |
| Zurich | Zurich Patch 7b or Zurich Patch 9 |
| Yokohama | Yokohama Patch 12 Hot Fix 1b or Yokohama Patch 13 |
| Hosted instances | Protected by ServiceNow since early April 2026 |
In addition, the update adds Guarded Script, a feature that limits sandboxed code from untrusted inputs, such as filter parameters, to a single simple expression. Because Guarded Script can block legitimate custom scripts, your platform team should test custom logic after the upgrade.
When Did Attacks Begin?
The window between disclosure and attack lasted only days. For that reason, these dates set the start of your log review.
| Date (2026) | Event |
|---|---|
| April 1 | Researchers report the flaw to ServiceNow |
| Early April | ServiceNow protects hosted instances within about 24 hours |
| June | Patches reach self-hosted customers and partners |
| July 13 | ServiceNow publishes KB3137947 and the CVE record |
| July 14 | Researchers publish a technical write up with a working sandbox escape |
| July 17 | First reported exploitation attempts against /assessment_thanks.do |
| July 18 | Researchers confirm a second gadget chain and a wider wave of attempts |
| July 20 | ServiceNow says it sees no evidence the activity relates to hosted instances |
What Can Attackers Do After Getting In?
The researchers’ proof of concept went well beyond running one command. For example, it showed that an attacker could create new accounts with admin rights and full read and write access to any table, which gives persistence that survives a patch. It also showed a path to shell commands on connected MID Servers, the proxies that link ServiceNow to on-premises networks.
Workflow data raises the stakes further. Tickets often hold passwords, network diagrams, incident notes, HR records and integration secrets. Consequently, a single ServiceNow RCE compromise can hand an attacker a map of the whole business and a bridge into the internal network.
How Strong Is the Evidence?
Most core ServiceNow RCE facts come straight from the vendor, yet several popular claims rest on single reports. Therefore, this table separates the confirmed points from the reported and open ones.
| Claim | Status |
|---|---|
| Unauthenticated code execution, CVSS 9.5, fixed releases | Confirmed by ServiceNow advisory KB3137947 |
| Hosted instances protected since early April | Confirmed by ServiceNow |
| Exploitation attempts from July 17 | Reported by one threat intelligence firm |
| A second gadget chain differs from the public proof of concept | Reported by the same firm |
| First captured request | A probe that tested code execution, not a destructive payload |
| Admin account creation and MID Server commands | Shown in the researchers’ proof of concept, not yet reported in the wild |
| Confirmed victims or data theft | None published |
| CISA KEV listing | Not found in our research |
What Are the Urgent ServiceNow RCE Fixes?
Order matters for every ServiceNow RCE fix, because a patch alone cannot remove an account that an attacker already created. Work through these seven steps from top to bottom.
Fix 1: Verify the Release and Patch
First, inventory every ServiceNow instance and record whether you host it yourself, a partner hosts it, or ServiceNow hosts it. Then upgrade any self-hosted or partner-hosted instance below a fixed release, and do not wait for the monthly maintenance window.
Fix 2: Restrict the Endpoint if You Cannot Patch Today
If a patch must wait, work with ServiceNow or your hosting partner to limit network reach to the affected pre-authentication endpoints. However, treat this step as a stopgap, since it does not remove the flaw.
Fix 3: Do Not Trust WAF Rules Alone
Next, remember that the in the wild gadget chain does not match the public payload. A WAF rule can cut noise, but only the patch closes the sandbox escape.
Fix 4: Hunt the Logs Back to July 13
After that, search for requests to /assessment_thanks.do since July 13, 2026. Focus on sysparm_assessable_type values that contain javascript:, Object.defineProperty, gs.include or Object.extendsObject, and treat any match as a likely compromise.
Fix 5: Audit Admin Accounts and Persistence
Meanwhile, compare every admin account in sys_user against approved users, service accounts and break glass access. Also look for unexpected scheduled jobs, new script includes, role changes and API token use, and disable anything you cannot explain.
Fix 6: Investigate Every MID Server
Likewise, review MID Server command logs, job history, outbound connections and credential use for the exposure window. Treat any MID Server that an exposed instance could reach as involved until the logs prove otherwise, and trim its network access to what it truly needs.
Fix 7: Run Incident Response When You Find a Match
Finally, if an internet facing instance stayed unpatched after July 13, open a formal investigation. Rotate credentials stored in tickets, integration accounts and MID Server configurations, and review Guarded Script results so the fix does not break critical workflows.
Which Indicators Should You Hunt For?
Load these reported indicators into your SIEM and web logs. Because ServiceNow RCE attackers can change infrastructure quickly, rely on the endpoint and string patterns more than the single IP address.
Why Do Workflow Platforms Draw Attackers?
Ultimately, ServiceNow holds the operational truth of a company: who owns each system, which incidents happened, and which vendors connect where. Our 2026 Mid Year Threat Pattern Report tracks the same push toward platforms that sit at the center of business processes, and our analysis of an AI PowerShell Active Directory attack shows how fast attackers turn that kind of access into control.
For that reason, self-hosted platforms need clear owners, version visibility and an emergency patch path. Our team runs penetration testing against web applications and workflow platforms, vulnerability assessments that find unpatched and exposed instances, and a PTaaS platform that keeps that testing continuous.
Key Takeaways
- CVE-2026-6875 lets an unauthenticated attacker escape the ServiceNow script sandbox through /assessment_thanks.do, with a CVSS v4.0 score of 9.5
- ServiceNow protected hosted instances in early April and shipped self-hosted patches through June before the July 13 disclosure
- Attempts began July 17, and attackers used a different gadget chain than the public proof of concept
- Admin account creation and MID Server commands appear in the proof of concept, so audit both even after patching
- Patch the ServiceNow RCE flaw first, then hunt logs back to July 13, audit sys_user and scheduled jobs, and review MID Servers
Frequently Asked Questions About the ServiceNow RCE Flaw
What Is the ServiceNow RCE Flaw CVE-2026-6875?
CVE-2026-6875 is a sandbox escape in the ServiceNow AI Platform that lets an unauthenticated attacker run code on a vulnerable instance. It scores 9.5 Critical under CVSS v4.0, and ServiceNow published its advisory, KB3137947, on July 13, 2026.
How Does CVE-2026-6875 Work?
An attacker sends a crafted request to the pre-authentication endpoint /assessment_thanks.do with a filter value that starts with javascript:. The platform evaluates that value, and a gadget chain then breaks out of the script sandbox and reaches code execution.
Are Hosted ServiceNow Instances Exposed?
ServiceNow pushed protections to hosted instances within about a day of the early April report. It also says it has seen no evidence that the July attacks relate to instances it hosts, so self-hosted and partner-hosted instances carry the most risk.
When Did Attacks Begin?
Researchers first saw exploitation attempts on July 17, 2026, four days after public disclosure, and confirmed a wider wave the next day. The first captured request worked as a probe that tested code execution rather than a destructive payload.
Why Do WAF Rules Fall Short?
Attackers reached the same code execution primitive through a different gadget chain than the public proof of concept. As a result, rules written against the published payload can miss real attacks, so the patch remains the only full fix.
Which ServiceNow Releases Fix the Flaw?
The fixed releases are Brazil EA and GA, Australia Patch 2, Zurich Patch 7b or Patch 9, and Yokohama Patch 12 Hot Fix 1b or Patch 13. ServiceNow says any self-hosted instance on an earlier release remains vulnerable.
What Should Teams Do After Patching?
Hunt for requests to /assessment_thanks.do since July 13, audit sys_user for new admin accounts, check scheduled jobs and script includes, and review MID Server command logs. If you find a match, treat the instance as compromised and rotate connected credentials.

