SECURITYCRITICALCVE-2026-21589

CVE-2026-21589: Atlassian Multiple Products Files or Directories Accessible to External Parties Vulnerability

Pre-auth path traversal flaw (CVSS 9.3) exposes files in 8 Atlassian Data Center products; CISA confirms active exploitation.

Dylan H.

Security Team

October 15, 2026
8 min read
CVE-2026-21589: Atlassian Multiple Products Files or Directories Accessible to External Parties Vulnerability

Actively exploited

Reported as exploited in the wild (e.g. CISA KEV). Patch or mitigate immediately.

Affected Products

  • Bitbucket Data Center — All Versions (fixed in 9.4.26, 10.2.8, 10.5.1)
  • Confluence Data Center — All Versions (fixed in 9.2.26, 10.2.19)
  • Jira Software Data Center — All Versions (fixed in 9.12.40, 10.3.26, 11.3.12)
  • Jira Service Management Data Center — All Versions (fixed in 5.12.40, 10.3.26, 11.3.12)
  • Bamboo Data Center — All Versions (fixed in 10.2.24, 12.1.12)
  • Crowd Data Center — All Versions (fixed in 6.3.7, 7.0.3, 7.1.7, 7.2.4)
  • Crucible — All Versions (fixed in 4.9.15)
  • Fisheye — All Versions (fixed in 4.9.15)

Overview

Atlassian has disclosed a critical, pre-authentication vulnerability — tracked as CVE-2026-21589 and classified under CWE-552 ("Files or Directories Accessible to External Parties") — affecting eight self-hosted Data Center and Server products: Bitbucket, Confluence, Jira Software, Jira Service Management, Bamboo, Crowd, Crucible, and Fisheye. The flaw allows an unauthenticated remote attacker to read specific files inside an affected application's web root, provided the attacker already knows the exact file name and path.

Atlassian published its advisory and fixed versions on October 5, 2026. Security researchers at watchTowr released a technical breakdown the following day, and exploitation attempts began within roughly 24 hours of that publication. On October 15, 2026, CISA added CVE-2026-21589 to its Known Exploited Vulnerabilities (KEV) catalog, confirming active exploitation in the wild and setting a federal civilian remediation deadline of October 18, 2026 — a three-day turnaround that reflects the severity of the issue.

Atlassian Cloud is not affected — the vendor-operated Cloud editions were already patched and Atlassian's investigation found no evidence of exploitation there. This advisory applies to self-managed Data Center and Server deployments only.


Technical Details

FieldValue
CVE IDCVE-2026-21589
CWECWE-552 — Files or Directories Accessible to External Parties
CVSS Score9.3 (Critical)
CVSS VectorCVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:N/VA:N/SC:H/SI:H/SA:H
Attack VectorNetwork (HTTP, path traversal)
AuthenticationNone required
Privileges RequiredNone
User InteractionNone
Enumeration RequiredAttacker must know the exact target file path — no directory listing or browsing
CISA KEV StatusAdded 2026-10-15; federal due date 2026-10-18
ImpactDisclosure of sensitive configuration files, credentials, and tokens stored in the web application root

How It Works

The root cause sits in atlassian-plugins-webresource, a shared library that Atlassian's Data Center and Server products use to serve static web resources such as images, scripts, and stylesheets. The library has a routing quirk: when it builds internal resource URLs, it encodes each forward slash in a resource name as a double colon (::), then decodes those double colons back into slashes immediately before loading the referenced file from disk.

That decode step happens after the framework's normal path-traversal defenses have already evaluated the URL, so an attacker can smuggle traversal sequences past the usual filters by hiding them inside the double-colon encoding. Once decoded, the resulting path is used to read a file from anywhere within the application's web root — without requiring a login, a valid session, or any user interaction.

The practical limitation is that this is not an arbitrary-enumeration bug: there's no directory listing, so an attacker needs to already know (or guess, from public documentation of default install layouts) the specific file path they want to read. In practice, that is a low bar — Atlassian's products ship with well-documented, largely static directory structures, and public proof-of-concept tooling (including a Python exploit script and a Nuclei template) emerged within days of disclosure, cataloguing high-value default file paths.

The impact varies by product, but the common thread is exposure of configuration files that frequently contain secrets:

  • Database connection strings and credentials
  • LDAP/directory-service bind credentials
  • API keys and integration tokens
  • Application configuration revealing internal topology

Particular Risk: Atlassian Crowd

Deployments using Crowd for centralized identity/user management face an outsized risk. The files reachable through this vulnerability in Crowd installations can include credentials sufficient for an attacker to create new user accounts and assign them to administrator groups — turning an information-disclosure bug into a path to full administrative takeover of every application federated through that Crowd instance.


Impact Assessment

Who Is At Risk

Any organization running a self-hosted (Data Center or Server) instance of the eight affected products, prior to the fixed versions, with network access to the application's HTTP interface. This includes:

  • Internet-facing Jira, Confluence, or Bitbucket Data Center instances
  • Internal instances reachable from a broader corporate network or VPN
  • Managed service providers hosting multi-tenant Atlassian Data Center environments
  • Any environment using Crowd as a central identity provider for other Atlassian tools

Potential Attack Chains

  1. Reconnaissance — Attacker identifies a target Atlassian Data Center instance and its product/version via banners, default pages, or fingerprinting.
  2. Pre-Auth File Read — Attacker sends a crafted request exploiting the double-colon path-traversal quirk to retrieve a known configuration file.
  3. Credential Extraction — Harvested database, LDAP, or Crowd credentials are parsed out of the retrieved file.
  4. Privilege Escalation / Lateral Movement — In Crowd deployments, harvested credentials are used to create or elevate accounts to administrator status; in other products, credentials may grant direct database or backend access.
  5. Full Compromise — Administrative access to Jira, Confluence, or Bitbucket enables data exfiltration, supply-chain tampering (for Bitbucket-hosted repositories), or deployment of further payloads.

Exploitation Status

This vulnerability is confirmed under active exploitation. CISA's October 15, 2026 KEV addition followed observed in-the-wild exploitation attempts that began within about a day of Atlassian's October 5 advisory. Atlassian products have a documented history of being targeted by ransomware operators and APT groups — Atlassian CVEs account for multiple prior KEV entries, several of which have been explicitly tied to ransomware campaigns. Organizations should treat this advisory with the same urgency as a confirmed, in-progress attack.


Mitigation

Immediate Actions

  • Patch now. Upgrade to the fixed version for your product:
    • Bitbucket Data Center: 9.4.26, 10.2.8, or 10.5.1
    • Confluence Data Center: 9.2.26 or 10.2.19
    • Jira Software Data Center: 9.12.40, 10.3.26, or 11.3.12
    • Jira Service Management Data Center: 5.12.40, 10.3.26, or 11.3.12
    • Bamboo Data Center: 10.2.24 or 12.1.12
    • Crowd Data Center: 6.3.7, 7.0.3, 7.1.7, or 7.2.4
    • Crucible: 4.9.15
    • Fisheye: 4.9.15
  • If immediate patching isn't possible, apply Atlassian's published interim mitigations:
    • A WAF or reverse-proxy regex rule blocking the double-colon path-traversal pattern (published in Atlassian's advisory)
    • A Tomcat RewriteValve configuration for Confluence, Jira Software, Jira Service Management, Bamboo, and Crowd
    • A urlrewrite.xml rule for Bitbucket
  • Restrict network exposure. Remove affected instances from the public internet, or limit access to trusted internal networks/VPN, until patched.
  • Rotate credentials stored in configuration files reachable from the web root (database passwords, LDAP bind credentials, API tokens) if there is any indication of prior exposure.

Detection Opportunities

  • Review HTTP access logs for requests containing double-colon (::) sequences in resource paths, or unusual encoded traversal patterns targeting static resource endpoints.
  • Look for requests to known sensitive file paths (configuration files, .properties, .xml, credential stores) served through web-resource or static-asset handlers.
  • In Crowd environments, audit recent user creation and group-membership changes for unexplained new administrator accounts.
  • Correlate any suspicious file-read activity with subsequent authentication events using previously unseen credentials.

Defence-in-Depth

  • Apply the principle of least privilege to service accounts running Atlassian applications so that even a successful file read yields minimal usable credentials.
  • Segment Crowd and other identity infrastructure from general application network access.
  • Maintain a regular patch cadence for Atlassian Data Center products — this is not the first and will not be the last high-severity Atlassian CVE.
  • Enable centralized logging and alerting for Atlassian application web tiers so anomalous request patterns are caught quickly.

Background: Atlassian as a Recurring KEV Target

Atlassian's self-hosted Data Center and Server products have repeatedly appeared on CISA's Known Exploited Vulnerabilities catalog, with past entries including both authentication-bypass and remote code execution flaws — several of which were subsequently weaponized by ransomware affiliates. The scale and criticality of Jira, Confluence, and Bitbucket deployments inside enterprise environments make them a persistent, high-value target: a single compromised instance often provides an attacker with source code, project data, credentials, and a pivot point into the broader internal network.

CVE-2026-21589 fits this pattern closely. It is unauthenticated, requires no user interaction, and — through Crowd specifically — offers a realistic path from simple file disclosure to full administrative compromise. Combined with public proof-of-concept tooling and confirmed in-the-wild exploitation, organizations running any of the affected products should treat patching as an emergency-change priority rather than part of a routine maintenance window.


References