Prompt Injection Risks Expose Data Through Atlassian Rovo

Researchers demonstrated two prompt injection techniques that could make Atlassian Rovo collect accessible data from Jira, Confluence, and connected services and transmit it to attacker-controlled servers. Atlassian fixed the one-click RovoBlast URL attack server-side on July 8, but the remediation status of a separate content-borne method reported by PromptArmor remains unconfirmed. Both attacks operate with the signed-in user’s existing permissions, and neither disclosure reports evidence of exploitation against real organizations. Administrators should limit Rovo access, tighten user and connector permissions, and avoid relying solely on the web-search setting to prevent data exfiltration.

Reference: thehackernews.com