Disclosed vulnerability
ServiceNow Now Assist: second-order prompt injection recruits privileged agents via agent-to-agent discovery
AIR-2025-0177 AIR record · 2025-11-servicenow-agent-discovery-injection
From the supporting source
I instructed a seemingly benign Now Assist agent to recruit more powerful agents in fulfilling a malicious and unintended task.appomni.com
The record
Disclosed
2025-11-19
Vendor
ServiceNow
Product
ServiceNow Now Assist AI Agents
Risk domain
Access control and permissions › Privilege escalation
Also touches
Data exfiltration channels, Confused deputy
System type
Generative model
Who acted
Agent acted — the Now Assist agent recruited privileged agents and emailed data out
Attack surface
Multi-agent system
Autonomy level
Semi-autonomous
Initial vector
Indirect prompt injection ticket
Impact type
Data exfiltration, Privilege escalation, Data integrity loss
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI03 – Identity and Privilege Abuse
- MITRE ATLAS
- AML.T0055 – Unsecured Credentials
- NIST AI RMF
- GOVERN 6.1
- EU AI Act
- Article 13
- ISO/IEC
- TR 24028; 42001; 23894
- AIUC-1
- B. Security
- Scope
- tool misuse
- Mitigation
- Apply least-privilege principle; Regular policy audits.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
ServiceNow instances running Now Assist agents with default team grouping and discoverability
Capability abused
Defense & remediation
Recorded defense
ServiceNow prompt-injection protection was enabled during the demonstration
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
none; ServiceNow confirmed the behaviors were intended and updated platform documentation instead
Taxonomy & references
Curator notes
AppOmni Labs (Aaron Costello). Published 2025-11-19. A low-privileged user who can only create tickets plants hidden instructions; a benign agent later reads them and recruits more privileged agents that act with the initiating admin's authority. ServiceNow 'confirmed these behaviors were intended'; this is a configuration/default-posture failure, not a patched bug, which makes it a standing risk rather than a closed one. Root causes cited: insecure LLM selection and default team-based agent grouping.
Cite as
Enkrypt AI. "ServiceNow Now Assist: second-order prompt injection recruits privileged agents via agent-to-agent discovery". AIR, AIR-2025-0177. https://agentincident.vercel.app/i/AIR-2025-0177/ (accessed 2026-10-19).