One click on a link just leaked your Jira, Confluence, and SharePoint. Atlassian's AI assistant had no guardrails.

Varonis just dropped RovoBlast at DEF CON 34.

A single crafted link injects attacker instructions directly into Atlassian Rovo's AI session.

No jailbreak. No permission bypass. No warning.

Rovo is an AI layer spanning Jira, Confluence, Bitbucket, Slack, Microsoft 365, and Google Workspace.

It also has an autonomous agent called ResearchAgent that can browse the web and post data externally.

One click seeds the prompt. ResearchAgent pulls internal data. Then pushes it to the open web.

Three proof-of-concepts were demonstrated: exfiltrating Confluence pages, Jira tickets, and SharePoint content containing personal data.

The organization ID in the URL can be left blank. Atlassian routes the request into the victim's default org with zero indication the session was seeded externally.

Rovo cannot be fully uninstalled. Organizations trying to remove the attack surface cannot eliminate Rovo's presence.

This is the same parameter-to-Prompt pattern Varonis found in Microsoft Copilot in January. It is now an industry-wide class of vulnerability.

Your CISO needs to audit every AI assistant with federated search and autonomous agent capabilities. Limit which systems Rovo can access. Disconnect unused integrations. Wall off legal, HR, and finance. Disable browsing agents and multi-step automation you do not actively use.

The lethal trifecta is real: private data access, untrusted content exposure, and external communication paths. Every enterprise AI assistant that combines these three is a breach waiting to happen.

Audit your AI assistant config today.

SOURCE: https://www.varonis.com/blog/rovoblast
VERIFIED: SecurityWeek (August 8, 2026), Varonis Threat Labs (August 7, 2026), The Hacker News (August 8, 2026)
SIGNAL: Atlassian Rovo connects to Jira, Confluence, SharePoint, and 50+ connectors. A one-click exploit that exfiltrates data across all of them with no guardrails is the most critical enterprise AI vulnerability disclosed this year.