Atlassian Rovo AI Flaw Could Leak Jira and Confluence Data

Security researchers have identified vulnerabilities in Atlassian Rovo, the company's AI assistant, that could potentially trick it into sending Jira and Confluence data accessible to an authenticated user to an attacker-controlled server.

One attack uses indirect prompt injection, where malicious instructions are hidden inside content processed by Rovo. Another, dubbed RovoBlast, abused a crafted Rovo URL to execute attacker-controlled instructions using the victim's permissions. Atlassian fixed the URL-based vulnerability server-side on July 8, 2026.

The findings highlight the growing security risks around AI assistants connected to enterprise data. Organizations should review Rovo access, limit unnecessary Jira/Confluence permissions and third-party connectors, and monitor unusual AI-driven data access or outbound requests. AI security should complement existing endpoint, application and data protection controls.

Secure Every Endpoint. Protect Every Byte with NPAV Data Loss Prevention (DLP)