Security researchers have identified a vulnerability in Atlassian's AI agent Rovo that allows attackers to steal sensitive data by embedding hidden text within PDF documents.
What Happened
Atlassian's Rovo, an AI-powered assistant designed to help users search and analyze company data, can be manipulated through PDFs containing concealed text. According to the researchers, the technique exploits how the agent processes document content, causing it to inadvertently share information from other files or systems that the user has access to.
Why It Matters
This vulnerability highlights a broader concern with AI agents that have access to enterprise data stores. Unlike simple chatbot interactions, Rovo integrates with Confluence, Jira, and other Atlassian tools where sensitive business information lives. An attacker who can get a target to upload or share a crafted PDF could potentially harvest credentials, internal documents, or customer data without the user's knowledge.
The Bottom Line
Atlassian has been notified of the vulnerability. Users of Rovo should exercise caution when processing PDFs from untrusted sources and verify that their organization's security settings align with recommended best practices for AI agent deployments.