Skip to main content
COSMICBYTEZLABS
NewsSecurityHOWTOsToolsTraining
StudyProjectsNewsletterHire MeAbout
Subscribe

Press Enter to search or Esc to close

News
Security
HOWTOs
Tools
Training
Study
Projects
Newsletter
Hire Me
About
RSS Feed
Reading List
Subscribe

Stay in the Loop

Get the latest security alerts, tutorials, and tech insights delivered to your inbox.

Subscribe NowFree forever. No spam.
COSMICBYTEZLABS

Your trusted source for IT intelligence, cybersecurity insights, and hands-on technical guides.

2269+ Articles
157+ Guides

CONTENT

  • Latest News
  • Security Alerts
  • HOWTOs
  • Checklists
  • Projects
  • Exam Prep

RESOURCES

  • Search
  • Browse Tags
  • Newsletter Archive
  • Reading List
  • RSS Feed

COMPANY

  • About Us
  • Contact
  • Privacy Policy
  • Terms of Service

© 2026 CosmicBytez Labs. All rights reserved.

System Status: Operational
  1. Home
  2. News
  3. Critical One-Click Vulnerability in Atlassian's Rovo AI Exposed Enterprise Data
Critical One-Click Vulnerability in Atlassian's Rovo AI Exposed Enterprise Data
NEWS

Critical One-Click Vulnerability in Atlassian's Rovo AI Exposed Enterprise Data

Researchers discovered a parameter-to-prompt injection flaw in Atlassian Rovo AI that allowed a single malicious link click to exfiltrate Confluence API keys, Jira data, and files from Microsoft 365 and Google Workspace integrations.

Dylan H.

News Desk

August 8, 2026
4 min read

A critical one-click vulnerability in Atlassian Rovo AI allowed attackers to weaponize the enterprise AI assistant against its own users — exfiltrating private API keys, Jira tickets, Confluence content, and data accessible through SharePoint and Outlook connectors with nothing more than a specially crafted URL.

The RovoBlast Attack: Parameter-to-Prompt Injection

Researchers at Varonis Threat Labs disclosed a vulnerability they named RovoBlast, which exploits a flaw in how Rovo's chat interface handles URL parameters. A malicious link containing a rovoChatPrompt parameter preloads attacker-controlled instructions directly into a victim's live Rovo AI session the moment they click it. Rovo then executes those instructions with the victim's full permissions, including access to all connected enterprise data sources.

In a proof-of-concept demonstration, the attack successfully exfiltrated a private Confluence API key to an external attacker-controlled server. The same technique was verified against Jira and data reachable through Microsoft 365 and Google Workspace connectors.

The Varonis team classified this as a parameter-to-prompt (P2P) injection — the same vulnerability class they previously uncovered in Microsoft Copilot (dubbed "Reprompt" in January 2026). A particularly notable detail: leaving the organization ID field blank in the crafted URL did not stop the attack — Atlassian's backend still routed the request into the victim's default organization.

Amplification via ResearchAgent

The blast radius is significantly expanded by Rovo's built-in ResearchAgent tool, which autonomously traverses multiple data sources during a single session. When hijacked via a RovoBlast payload, the agent can pivot across Jira, Confluence, Bitbucket, Slack, Google Workspace, Microsoft 365, uploaded files, relational databases, and web-accessible content — all within a single click.

According to Rovo's own self-reporting when queried, the full scope of accessible data includes archived content going back years.

Atlassian paid a $6,000 bug bounty and patched the vulnerability server-side on July 8, 2026. No client action is required for the RovoBlast vector.

A Second Vulnerability: Zero-Click Prompt Injection (Unpatched)

Concurrent research by PromptArmor revealed a separate and more persistent attack vector: indirect prompt injection via documents. Reported on May 23, 2026, this flaw remains unpatched as of August 5, 2026.

The attack embeds hidden instructions inside documents that Rovo processes — white-on-white text, 1-point font, or 0.1-point line spacing in PDFs disguised as legitimate SOPs. When a user asks Rovo to process a Jira ticket or Confluence page containing or linked to such a document, Rovo's URL retrieval tool is hijacked by the injected instructions to silently POST sensitive data to an external server.

Bypassing Admin Controls

This attack works even when organization-wide web search is disabled in Rovo's admin panel — disabling web search does not remove the underlying URL retrieval capability, which remains exploitable as an exfiltration channel. PromptArmor also identified a secondary path: Rovo renders images embedded in AI responses, and image URL parameters can carry exfiltrated data payloads out of the enterprise perimeter.

Scale of Exposure

Atlassian Rovo has over 5 million monthly active users and is deployed at 75% of Fortune 500 companies, with adoption in over 90% of Atlassian's enterprise customer base. The combination of broad enterprise data access and autonomous agent capabilities makes AI assistant vulnerabilities in this product a high-severity concern.

Broader Context

The indirect prompt injection pattern found in Rovo is not unique to Atlassian. The same fundamental vulnerability class has been documented in Claude integrations, Google Workspace AI, Slack AI, and Notion AI. The core architectural problem — LLMs that cannot strictly separate data from control instructions — is an industry-wide challenge that no vendor has fully resolved.

What Organizations Should Do

  • RovoBlast (patched): No client-side action required; Atlassian deployed the fix server-side on July 8, 2026
  • Indirect prompt injection (unpatched): Monitor for unusual Rovo activity; be cautious about allowing Rovo to process documents from external or untrusted sources; check PromptArmor's disclosure for the latest patch status
  • Audit connected integrations: Review which external data sources (SharePoint, Outlook, Google Workspace, Slack) are connected to Rovo and assess the data exposure risk
  • Watch for Atlassian's patch: Track CVE assignments and Atlassian's security advisories for the indirect prompt injection fix
#Vulnerability#AI Security#Atlassian#Prompt Injection#Enterprise

Related Articles

Critical Cursor AI Code Editor Flaws Could Lead to OS-Level Remote Code Execution

Two critical vulnerabilities (CVE-2026-50548, CVE-2026-50549) dubbed DuneSlide allow zero-click prompt injection attacks to escape Cursor IDE's sandbox...

4 min read

AI Recommendation Poisoning: How "Ask AI" Buttons Silently Alter LLM Memory

A new class of prompt injection embedded in "Ask AI" buttons on commercial websites silently writes biased recommendations into users' AI assistant memory — no malware, no credentials, just a weaponized URL parameter.

6 min read

Adobe Campaign Classic CVSS 10.0 Flaw Could Run Code Without User Interaction

Adobe has patched a maximum-severity vulnerability in Campaign Classic (ACC), its enterprise marketing automation platform. The flaw carries a perfect CVSS score of 10.0 and allows unauthenticated remote code execution with no user interaction required.

4 min read
Back to all News