RovoBlast Vulnerability Exposes Enterprise Data via One-Click Attack

Atlassian's Rovo AI assistant was found vulnerable to a one-click prompt injection attack, dubbed RovoBlast, which allows attackers to inject malicious instructions into authenticated user sessions. This vulnerability was disclosed by Varonis Threat Labs at DEF CON 34 and affects Rovo's integration with services like Jira, Confluence, and Slack. The attack can exfiltrate sensitive data without user approval, as Rovo's browsing agent autonomously retrieves and posts internal content externally. The flaw was reported to Atlassian, which has since deployed a fix on July 8, 2026. However, Rovo's inability to be fully uninstalled means organizations still face risks. Researchers recommend limiting Rovo's access and monitoring its activity to mitigate potential threats.

Key Points: • RovoBlast allows one-click data exfiltration from Atlassian's Rovo AI assistant. • The vulnerability affects multiple platforms, including Jira, Confluence, and Slack. • Atlassian has patched the vulnerability, but Rovo's persistent presence poses ongoing risks.