ServiceNow CVE-2026-18883
Indirect Prompt Injection in Email Response AI Agent (Now Assist for CSM)
Risk & exploitation
Detection seeds: none indexed — public detection content (Sigma) that references this id; a seed is a starting point for a rule, not a verified fingerprint.
Disclosed 3 days ago, so the exploitation signals this score leans on — EPSS, CISA KEV and public-exploit indexing — have not had time to mean anything yet. Priority follows CVSS severity until they do, rather than reporting their absence as safety. EPSS (FIRST) estimates a low likelihood of exploitation in the next 30 days (2026-11-20). Priority blends CVSS, EPSS, KEV and public-exploit availability — operational prioritization, not a NIS2/DORA reporting determination.
Summary
What it is. An indirect prompt injection in the Email response AI agent of Now Assist for Customer Service Management. Text the agent reads as input, such as an inbound email or the case it answers, can carry instructions the agent then follows. The attacker needs no login, only a way to get text in front of the agent. ServiceNow scores it 7.1 (CVSS 4.0), high. The CVE Record assigns no CWE; the class is the one CWE-1427 describes, improper neutralization of input used for LLM prompting.
Are you exposed? Only if the Customer Service Management AI agent collection plugin is installed, at a version below 6.1. KB3235180 states that an instance without the app is not affected. The fix ships in the plugin, not in a platform patch, so your release and patch level do not answer this one: check the plugin's version in Application Manager.
Why it matters beyond this CVE. The Email response agent drafts and sends email from source-record details, so an injected instruction can reach a send action. Patching closes this flaw; it does not answer whether agents that read untrusted text can take actions with no human approving them. That is a configuration question for every agent on the instance.
Exploitation signals. ServiceNow reports no evidence of malicious exploitation. No public exploit code is tracked for this CVE and it is not in CISA KEV.
Affected releases & fixed-in
| Release | Fixed in |
|---|---|
| Customer Service Management AI agent collection | 6.1 (CVE Record + KB) |
Source & attribution
Fixed-in verified 2026-11-20 against ServiceNow KB3235180 ('[Security Advisory] November 2026 CVE Advisory Notification', published 2026-11-19, read by a human) and cross-checked against the ServiceNow-authored CVE Record (CNA shortName: SN), https://cveawg.mitre.org/api/cve/CVE-2026-18883. KB3235180: customers with the plugin installed should be on version 6.1 or higher, and an instance without the app is not affected. The CVE Record names the affected product 'SR - CSM - Customer Service Management AI agent collection' (the Store listing); affected[].release carries the installed plugin's name, 'Customer Service Management AI agent collection', as ServiceNowDocs spells it (https://raw.githubusercontent.com/ServiceNow/ServiceNowDocs/96d32129bb43ac72057be47de4ffd83cf128813c/markdown/customer-service-management/c_PhoneCommunication.md and https://raw.githubusercontent.com/ServiceNow/ServiceNowDocs/96d32129bb43ac72057be47de4ffd83cf128813c/markdown/customer-service-management/now-assist-for-csm/csm-live-agent-assist.md), so an installed-app inventory can match it.
Regulatory mapping
- NIS2 Art.21§2(e)
- NIS2 Art.21§2(i)
- NIS2 Art.23§1
- DORA Art.9§4(c)
- DORA Art.9§4(f)
- DORA Art.19§1
- ISO A.8.8
- ISO A.5.15
- ISO A.8.2
Decision support, not a reporting determination.
References
Data: GitHub Advisory Database (CC-BY 4.0), NVD, the CISA KEV catalog, FIRST EPSS, Exploit-DB, and Nuclei templates.