Disclosed vulnerability
Atlassian Rovo: zero-click Jira/Confluence exfiltration via a poisoned uploaded file, bypassing web-search controls
AIR-2026-0185 AIR record · 2026-08-atlassian-rovo-exfiltration
From the supporting source
Atlassian's Rovo AI is susceptible to zero-click data exfiltration via indirect prompt injection, bypassing organization-level web search controls.promptarmor.com
The record
Disclosed
2026-08-05
Vendor
Atlassian
Product
Atlassian Rovo
Risk domain
Privacy › Data exfiltration channels
Also touches
Confused deputy
System type
Generative model
Who acted
Agent acted — Rovo searched Jira/Confluence and exfiltrated via the poisoned file
Attack surface
Enterprise assistant
Autonomy level
Semi-autonomous
Initial vector
Indirect prompt injection document
Impact type
Data exfiltration
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI02 – Tool Misuse and Exploitation
- MITRE ATLAS
- AML.T0024 – Exfiltration via AI Inference API
- NIST AI RMF
- MAP 4.2
- EU AI Act
- Article 10
- ISO/IEC
- TR 24028; 23894
- AIUC-1
- A. Data & Privacy
- Scope
- tool misuse
- Mitigation
- Monitor network traffic for anomalous patterns; Restrict egress destinations.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
Rovo tenants; any data the agent can reach in Atlassian and via connectors
Capability abused
Defense & remediation
Recorded defense
organization-level web search controls in Rovo
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
none; unresolved at publication
Taxonomy & references
Curator notes
PromptArmor. Disclosed to Atlassian 2026-05-23, acknowledged 2026-05-25, follow-ups 2026-06-04 and 2026-07-29, published 2026-08-05 with the note 'Rovo remains vulnerable as of the release of this article.' Distinct from and concurrent with Varonis's RovoBlast; two independent injection paths into the same assistant.
Supporting sources
Cite as
Enkrypt AI. "Atlassian Rovo: zero-click Jira/Confluence exfiltration via a poisoned uploaded file, bypassing web-search controls". AIR, AIR-2026-0185. https://agentincident.vercel.app/i/AIR-2026-0185/ (accessed 2026-10-19).