PDF Hijacks Atlassian's Rovo AI Agent With No Trace Left

A line of white text on a white background inside an ordinary PDF — and Atlassian's Rovo AI agent silently exfiltrates corporate data: Jira tickets, Confluence documents, architecture specs — straight to an attacker's server. Security firm PromptArmor discovered the attack: no user confirmation, no trace in the chat — the agent simply executes the hidden command. Atlassian was notified on May 23 and has not shipped a patch in 2.5 months.

AI agents with access to enterprise systems are becoming a new threat class: the more autonomous the agent, the easier it is to hijack. Prompt injection — embedding hidden instructions via documents — has become a standard attack vector against enterprise AI, and the window between discovery and remediation is stretching into months.