Rovo Flaw Lets Attackers Exfiltrate Jira/Confluence Data via Prompt Injection

Security researchers found that Atlassian's Rovo assistant can be tricked into sending Jira and Confluence data to attackers through attacker-controlled prompts and a malicious URL parameter. Two independent reports (PromptArmor and Varonis Threat Labs) detail a content-borne prompt injection path and a one-click link path, with Atlassian fixing the link-based flaw on July 8, 2026; the content-borne path’s status remained uncertain as of Aug 8, 2026. Exfiltration occurs within the victim’s signed-in permissions, and admins can mitigate by restricting Rovo usage by app/group or disabling Rovo features. No CVEs have been issued. Organizations should tighten app scopes and permissions rather than relying on the web-search toggle as a security boundary.