Four flaws in the ServiceNow AI Platform need no login to exploit

ServiceNow said on 27 August 2026 that four vulnerabilities in the ServiceNow AI Platform have been fixed. Three of them carry a CVSS score of 10.0, and the fourth carries 8.7. CVE-2026-6876 lets code get out of the sandbox and run freely, CVE-2026-18885 allows code execution through code injection, CVE-2026-18886 allows privilege escalation through improper access control, and CVE-2026-74820 lets an attacker run arbitrary SQL statements that could expose or alter data in the instance. All four can be exploited remotely without authentication, without any user interaction, and in low complexity attacks.

The company has updated the instances it hosts itself, and has published fixes for partners and for customers who run the platform on their own. Affected releases sit in the Xanadu, Yokohama, Zurich and Australia families, and the advisory lists which patch levels are fixed in each of them. ServiceNow states that it is not aware of malicious exploitation.

What this means for you if you run ServiceNow yourself

The division of labour is the heart of this story. If the vendor hosts your instance, the work is already done. If you run it yourself, the work has not started until you have looked up your own release in the advisory and compared the patch level. Our assessment is that this is where most organisations stall, because the responsibility was never written down anywhere.

ServiceNow is rarely a side system. The platform holds case handling, access requests and change management, so anyone able to run SQL against the instance sees large parts of the business at once. That nobody has observed exploitation yet is good news today and no guarantee next week. Four vulnerabilities that require no login rarely stay untouched for long.

Berigo recommends

• Look up which release and patch level your instance actually runs, and hold it against the ServiceNow advisory.
• Settle in writing whether you or the vendor installs security updates on each cloud platform you use.
• Feed ServiceNow security advisories into ordinary vulnerability management, so they do not sit in an inbox.
• Restrict who can reach the administrative interface, and log access to it.
• Review which integrations hold privileges into the platform, since a compromised instance reaches them too.