Finding leaked credentials in Jira and Confluence
In this guide
A webhook token in a Jira comment, a database password in a Confluence runbook, a cloud key in a screenshot attached to a ticket: credentials land in Atlassian products because people paste what they are debugging. Native search can find some of that text. It misses other places that still hold the same value.
LeakRadar is a secret and personal data scanner for Jira and Confluence Cloud from Radrly. This guide is a map of where leaked credentials hide, what native search covers and misses, and the rotate-first order to clean them up. The app comes in the second half. Advanced-only features are labelled (Advanced).
Where credentials show up
In this guide, credentials means secrets people paste into tickets and pages: API keys, cloud, code, CI and AI provider tokens, passwords, private keys and connection strings. Personal data such as card numbers or national IDs is a related problem, covered in the DLP for Jira and Confluence guide. The search limits below apply to both.
Typical locations are Jira summaries, descriptions and comments, Confluence pages, blog posts and comments, field history, older page versions, attachments (including HAR files) and text inside images or scanned PDFs.
What native Jira search covers
JQL text ~ searches the current text of summaries, descriptions, comments and text fields. It is a word search with stemming. It has no regular expressions and no checksums, so a labelled word such as password is easier to find than a key pasted with no label. It does not search the History tab and it does not read attachment contents. A clean JQL result tells you the current text is clean, and nothing more.
A five-search cookbook, with what each query catches and misses, is in API keys and passwords pasted into Jira.
What native Confluence search covers
Confluence site search looks at the current text of pages and blog posts. Editing a page leaves older versions in the page history, and anyone with view access can open that history and compare versions. A space admin can delete older versions one at a time. The current version always stays.
Comments and attachments need a separate look. Search of the current page body does not prove that a footer comment, an inline comment or a file attached to the page is clean.
Places native search misses
These places often still hold a credential after the current text looks clean:
- The Jira History tab, which keeps old values of summary, description and environment for anyone who can browse the work item.
- Notification emails sent when a comment or description was added. Deleting the comment does not recall those emails.
- Older Confluence page and blog post versions.
- Attachment contents, including HAR headers and text in screenshots.
If you already edited a key out of a ticket, here is where it still lives walks through those leftovers in more detail, including what Atlassian Guard Premium can redact.
The order to follow
1. Rotate or revoke the credential at its provider. 2. Redact the current text in Jira or Confluence. 3. Purge older Confluence versions that still contain it. 4. Check Jira history and restrict the work item if the value is still readable there. LeakRadar cannot check whether a secret is still active, because that would need a call outside Atlassian. Treat every confirmed credential as live.
Where LeakRadar fits
LeakRadar scans both products on the same licence: Jira summaries, descriptions, comments, the latest 20 history entries of summary, description and environment, and attachments; Confluence pages and blog posts, footer and inline comments, up to 15 older versions, and attachments. Text inside images and scanned PDFs is read with OCR, so screenshots count too. The first 10 files of an item are scanned, up to 5 MB each.
Built-in secret detectors use prefixes, validators, checksums, entropy and keyword context. They cover AWS access and secret keys, Google Cloud and Azure secrets, GitHub, GitLab and Atlassian tokens, npm, PyPI and Docker tokens, Slack tokens and Slack app-level tokens, Stripe and many other provider tokens, OpenAI and Anthropic keys, Hugging Face, Linear, Sentry, Fly.io, PlanetScale, Doppler and Terraform Cloud tokens, JSON Web Tokens and password assignments, plus PEM private keys and database or broker URLs that carry credentials. You can add custom detectors (up to 20 in Standard, up to 100 in Advanced) and turn detectors or a framework pack on or off.
A hit is definite when a format check such as a checksum or a known token prefix confirms it, or ambiguous otherwise. Ambiguous hits can go to Atlassian-hosted Forge LLMs, with every detected value replaced by a type, length and character set placeholder. The AI never hides a definite credential. LeakRadar stores a masked snippet, a keyed hash and the location, never the full value, and groups the same value found in several places into one finding.
From a finding you can assign an owner, follow the rotate-first playbook, redact the current text in place with [REDACTED by LeakRadar], purge older Confluence versions whose body still matches the keyed hash, delete an attachment or replace a HAR with a sanitised copy, and restrict access with a Jira security level or a Confluence read restriction. Apps cannot remove Jira field history. Confirmed history findings can be handed to Atlassian Guard with the Guard handoff CSV on Audit > Evidence.
What this scan does not prove
A clean scan does not cover content outside the saved scope, detectors you turned off, Jira custom fields other than summary, description and environment, video and audio, whiteboards, databases and live docs, or images over the monthly OCR limit (2,000 images per UTC month on Standard; no monthly cap on Advanced). Personal spaces and archived projects stay out until you turn them on. LeakRadar publishes no precision or recall figures yet. Rotate every confirmed credential anyway.
Quick answers
Can native search find leaked credentials in Jira and Confluence?
Only partly. Jira JQL and Confluence search look at current text. They do not read the History tab, attachment contents, text in images, or older Confluence page versions.
What should I do first when I find a leaked credential?
Rotate or revoke the credential at its provider before you edit any text. Then redact the current content, purge older Confluence versions that still contain it, and check Jira history.
Does LeakRadar store the full credential?
No. LeakRadar keeps a masked snippet, a keyed hash and the location. The full value is never written to storage or logs.