Hello Atlassian community
,
Keeping your teams' work secure shouldn't slow them down. At Team '26 Europe, we introduced new Atlassian Guard capabilities that give you more control over your data, more confidence in your AI adoption, and faster ways to act when something looks wrong. Here's what's available now and what's ahead.
Why Guard, and why now
Many organizations are moving fast with AI. They're deploying agents, connecting outside data sources and using AI search to find context quickly. As data moves more freely, confidential information is more likely to be accessed, shared or misused. For regulated and complex organizations, governance concerns can slow AI adoption down.
Guard is the advanced security and data governance layer for Atlassian. With Guard, security and IT teams can protect sensitive data, maintain governance standards and manage risk at scale.
At Team '26 Europe, we shared more capabilities that help you:
- Discover and remediate sensitive data: Scan all your existing Jira and Confluence content review the findings in Atlassian Administration, and redact in bulk. Only the confidential details are removed, so your teams keep the context they need to get work done.
- Prevent sensitive data from building up again: Set information protection policies that automatically redact high-risk data the moment it appears. You can also flag it as a violation for review, so you can focus on the alerts that matter most.
- Govern Rovo interactions: Choose what types of data are redacted or blocked in Rovo Chat prompts and responses. You can extend the same protection to connected sources like SharePoint and Google Drive, so your teams can bring more context into AI with confidence.
- Control Atlassian MCP access by classification: Use data security policies to control whether content can be accessed through the Atlassian MCP server based on its classification level. Confidential data stays protected even as AI tools and agents connect to your Atlassian work.
Top 5 new capabilities available now:
- Full-site scanning (Guard Premium): Scan existing Jira and Confluence content across specific sites or your whole organization. This finds legacy sensitive data, like an account number, password, or email buried in a five-year-old ticket, so you can clean it up at scale.
- Attachment scanning (Guard Premium): Guard scans files and images (PDFs, spreadsheets, screenshots) when they're uploaded or updated. If it finds sensitive information, admins can keep or delete the attachment.
- Findings (Guard Premium): One dashboard in Atlassian Administration for all data Guard finds, from both full-site scans and ongoing monitoring. You can mark false positives and redact findings one at a time or in bulk. Redaction removes only the confidential details, so the rest of the page or ticket stays usable.
- Information protection policies (Guard Premium):Set policies to automatically take action when Guard finds confidential data based on your organization's requirements. You decide what to look for, where to look, and whether to automatically redact it or flag it for review, turning one-off cleanups into proactive protection.
- Atlassian MCP server control (Guard Standard and Premium): Decide whether AI agents and assistants can read your Jira and Confluence content through the Atlassian MCP server, across your whole organization, specific sites, or individual spaces and projects. With Guard Premium, you can also control access by classification level. For example, you can block AI agents from reading anything labeled highly restricted, like financial information or M&A plans, without changing how your teams collaborate.
Top 5 capabilities coming soon:
- Rovo Chat security (Guard Premium): Set rules to block or redact sensitive data, like API keys or bank details, across both user prompts and responses.
- Connector data security (Guard Premium):Guard stops sensitive data from third-party connectors, like Google Drive, from being ingested into Atlassian. If a document in a connected source contains sensitive data, Guard prevents Rovo from indexing or surfacing it.
- Exact data match (Guard Premium):Detect information unique to your organization, such as employee IDs, and Guard looks for exact matches across Jira and Confluence, greatly reducing false positives.
- Semantic detections (Guard Premium):Guard detects sensitive content from its context and classifies it automatically based on your admins' rules. It catches confidential topics like M&A plans or unreleased roadmaps, and you can link classifications to policies that restrict MCP access and exports.
- In-line prevention (Guard Premium):Guard intercepts and stops sensitive data at the point of entry in Jira and Confluence, preventing out-of-policy data from being stored.
💬 Top 5 questions we heard at the booth
We loved talking with so many of you in person in Amsterdam. You asked a lot of great questions, so we're sharing the ones we heard most. Whether you stopped by the booth or you're catching up now, here are the answers.
| Question | Answer |
|---|
1 | We've had sensitive data sitting in Jira and Confluence for years. Can Guard help us clean it up? | Yes, and this was the question we heard most! Full-site scanning searches your existing Jira and Confluence content, across specific sites or your whole organization. It finds the account numbers, passwords and personal data buried in old tickets and pages. Everything it finds goes into Findings, where you can review it and redact it one item at a time or in bulk. Redaction removes only the confidential details, so the rest of the page or ticket stays usable. |
2 | Cleaning up once is great, but how do we stop it from piling up again? | That's what information protection policies are for. You decide what Guard looks for, where it looks and what happens next. Guard can redact high-risk data automatically the moment it appears, or flag it as a violation for review. Attachment scanning extends this protection to files and images like PDFs, spreadsheets and screenshots whenever they're uploaded or updated. Coming soon, in-line prevention will stop out-of-policy data at the point of entry. |
3 | We get a lot of alerts, and some are false positives. How does Guard help us with this problem? | You can mark false positives directly in Findings and scope policies to the sites, spaces and projects that matter most. You can also choose whether each policy redacts automatically or flags data for review, so your team spends its time on the alerts that need a human. Coming soon, exact data match will look for information unique to your organization, such as employee IDs, which greatly reduces false positives. Semantic detections will use context to find confidential topics like M&A plans, even when there's no predictable pattern. |
4 | We want to use Rovo, but we need our data cleaned up and protected first. Where should we start? | We heard this a lot, especially from regulated organizations. A great first step is a full-site scan. It shows how much sensitive data you have, and you can clean it up before you roll out AI more broadly. From there, Atlassian MCP server controls let you decide whether AI agents and assistants can read your content. With Guard Premium, you can block access by classification level. Coming soon, Rovo Chat security will block or redact sensitive data in prompts and responses. Connector data security will stop sensitive data from sources like Google Drive from being indexed or surfaced by Rovo. |
5 | Our sensitive data doesn't always fit a standard format. Can Guard handle what's unique to us? | Yes. Guard's detections go beyond simple pattern matching, and you can tailor them to your organization's requirements. Coming soon, exact data match and semantic detections will help you protect identifiers and confidential topics specific to your business. |
👉 What's next
To learn more about Guard, check out our latest guide, or start a free trial.
Have questions about keeping things secure or rolling out AI safely? Drop a comment below!
Please note that features marked as "coming soon" are currently in development, and their timing and functionality may change.
