Varonis Threat Labs uncovered RovoBlast, a one-click vulnerability in Atlassian’s Rovo AI assistant that let attacker-controlled instructions be injected into live AI sessions through a crafted link. The flaw could expose data from Jira, Confluence, Bitbucket, Slack, Microsoft 365, Google Workspace, and other connected systems before Atlassian fixed it. #RovoBlast #Atlassian #Rovo #Varonis #DEFCON34
Keypoints
- RovoBlast let a crafted URL seed attacker instructions into Rovo’s chat session.
- The issue relied on parameter-to-prompt injection through the rovoChatPrompt URL parameter.
- Rovo’s autonomous ResearchAgent could pull internal data and leak it to the web.
- Proof-of-concepts showed exfiltration from Confluence, Jira, and SharePoint.
- Atlassian patched the flaw after responsible disclosure from Varonis Threat Labs.