ServiceNow patches three CVSS 10.0 flaws, including unauthenticated code injection

ServiceNow ships fixes for three maximum-severity flaws

ServiceNow has patched three vulnerabilities that each carry a perfect 10.0 CVSS score and can be triggered remotely without any authentication, together with a fourth, lower-rated sandbox escape in its AI Platform. According to SecurityWeek, whose findings are summarised in a dev.to write-up, the fixes came through ServiceNow's August 2026 security advisory, and the company said it had seen no sign of active exploitation at the time of publication.

The vulnerabilities in detail

CVE-2026-18885 is a code injection flaw in the GraphQL Composite Data API. It lets an attacker who can reach an exposed instance over HTTP or HTTPS execute code and read or modify the data held there, with no credentials and no user interaction involved.

CVE-2026-18886 concerns access control in the processing of configuration image uploads. Per the dev.to analysis, an unauthenticated attacker can abuse it to create and alter arbitrary data and then escalate privileges on the platform.

CVE-2026-74820 is a SQL injection that manifests in ORDER BY clauses built against dynamic schemas, opening the door to arbitrary SQL execution on the database that sits underneath the instance.

Alongside these, CVE-2026-6876 — rated 8.7 — is a sandbox escape in the ServiceNow AI Platform that also results in unauthenticated code execution within the Now Platform.

Patch availability

ServiceNow has already rolled out the updates to instances it hosts itself. Self-hosted deployments must apply hotfixes manually, and the vendor supplies them across the Xanadu, Yokohama, Zurich and Australia release series.

How an attack could unfold

The attack path described by dev.to starts with a crafted request aimed at an externally reachable instance. Exploitation succeeds when three conditions line up: the target is reachable over the network, the relevant hotfix is not installed, and no web application firewall or API control intercepts the malicious request.

If any of the flaws land, the attacker can execute arbitrary code, create and modify instance data, escalate privileges and tamper with approval workflows. The SQL injection adds arbitrary database queries, and because ServiceNow instances typically hold integration credentials, a single compromise can spread outward to connected external systems.

What administrators should watch for

Warning signs include unusual GraphQL traffic, configuration image uploads nobody expected, database errors mentioning ORDER BY clauses, and data or permission changes traced back to sessions that never authenticated. After a compromise, defenders may also spot newly created high-privilege users, unexpected role assignments or integration credential usage, and outbound traffic toward unknown destinations.

The recommended response is straightforward: install the hotfix and confirm the patched version, restrict external exposure of administrative, API and upload functions to only the sources that genuinely need them, enable auditing for GraphQL activity, file uploads, SQL errors and permission changes, and keep integration credentials on a least-privilege footing.

For self-hosted environments, endpoint telemetry adds another layer: suspicious child processes of the ServiceNow application, unexpected file creation and script execution are all worth reviewing when investigating a possible incident.

Why it matters

ServiceNow instances tend to sit at the crossroads of HR, procurement and financial approvals, and they hold the credentials that tie those workflows to outside systems. An unauthenticated, no-interaction exploit rated 10.0 on an internet-facing instance is about as bad as an enterprise attack surface gets, and the blast radius extends well beyond the platform itself. With no exploitation observed so far, administrators have a window to patch before proof-of-concept code or attacker interest changes the picture — but that window should be treated as short, and self-hosted deployments left unpatched are the obvious first targets.