Atlassian Rovo data breach exposes Jira and Confluence files to hackers

Two independent security firms have documented an Atlassian Rovo data breach that enables a signed‑in user to be tricked into exporting Jira or Confluence content to a hostile server.

How the Rovo assistant can be weaponized

The attack hinges on malicious instructions concealed in files that Rovo parses. PromptArmor, an AI‑security specialist, demonstrated that a crafted document can embed a payload which Rovo silently executes, pulling any project data the user can view.

In a separate proof‑of‑concept, researchers from Reuters showed a different injection vector that remains unpatched, allowing exfiltration of up to ↓ 500 KB per request.

“Rovo’s trust model assumes content is benign, a premise that these findings shatter,” one analyst warned.

Only the PromptArmor route has been confirmed patched; the alternative pathway still poses a significant risk for enterprises that rely on Atlassian’s collaboration suite.