critical Cve 2026 21589 · Exploits

Atlassian Patches Unauthenticated File Read CVE-2026-21589 Across Eight Data Center Products

Data graphic: ThreatFrontier cover for the Atlassian Data Center pre-authentication file read, CVE-2026-21589, with a large "9.3" CVSS 4.0 Critical score, "8 products" affected and "None required" authentication.
AK

Threat intelligence editor · Published Oct 7, 2026 · Updated Oct 15, 2026, 8:46 PM EDT

Atlassian's CVSS 9.3 flaw lets anyone read files from eight Data Center products' web root without logging in. Fixed versions, mitigations and what to rotate.

Update (16 Oct 2026): now on CISA KEV, with exploitation widespread

CISA added CVE-2026-21589 to its Known Exploited Vulnerabilities catalogue on 15 October (catalogue version 2026.10.15, which we read directly). The KEV entry sets a due date of 18 October and flags forensic triage as required under BOD 26-04; its ransomware-use field is "Unknown". Exploitation was reported well before that. On 7 October BleepingComputer reported that security firm Previdian saw attempts against CVE-2026-21589 on its honeypot network within two hours of watchTowr publishing its technical research and proof of concept, and watchTowr's FAQ lists exploitation as "In The Wild". Also on 7 October, VulnCheck added the CVE to its own KEV list, citing exploitation activity targeting Bamboo Data Center, according to Infosecurity Magazine (VulnCheck's dashboard links to Previdian as a source). Bamboo is the only product named as targeted in the sources we read, but that may reflect how the CVE is catalogued: Previdian files it under "Atlassian / Bamboo Data Center", the first product in the CVE record, and the public scanning tools it lists target Jira, Confluence and Bitbucket. No source says which products' instances were actually probed or compromised, so do not read this as meaning only Bamboo is at risk.

Scale: Previdian's CVE page now counts 607 exploitation attempts from 65 unique IP addresses in 20 countries, first seen on 6 October and last seen on 15 October. Its daily chart peaks at 262 attempts on 7 October, falls to single digits by 12 October, then rises again to 30 and 28 on 14 and 15 October; Previdian also lists Shadowserver as attesting exploitation on 14 October. Those are one honeypot vendor's sensor counts, not victim counts, and the figure has grown since its early reading of three IP addresses (38.60.157[.]86, 146.70.187[.]234 and 159.26.119[.]225) on 7 October. Previdian's page also lists a Nuclei template for automated scanning (detected 7 October) and nine GitHub repositories claiming to hold proof-of-concept code, created between 6 and 9 October (one is watchTowr's own), so roughly ten public tools exist. One, created on 6 October, describes itself as a full exploit with a mass-scan mode. Previdian labels all of these unverified and possibly malware-laden; we have not run or checked any of them.

What the public research enables: in Jira deployments integrated with Crowd, the file read can return WEB-INF/classes/crowd.properties, which holds the Crowd application credentials in plaintext. Those credentials can then be used through Crowd's API to create a Jira administrator account, provided Crowd is reachable from the attacker's position and the application has sufficient permissions. watchTowr notes a Crowd IP allow-list makes this much harder. This is the researchers' demonstrated chain; no source we read says it has been used against a real victim.

If you run a Crowd-integrated deployment and were reachable and unpatched, rotate the application credentials in crowd.properties now, review Crowd and Jira for unexpected new administrator accounts, and patch or apply Atlassian's mitigations. Blocking source IP addresses is not a fix: Previdian's list has grown from three addresses to 65, and it comes from one vendor's sensors.

What happened

On 5 October 2026 Atlassian disclosed CVE-2026-21589, an arbitrary file access flaw that lets an unauthenticated attacker read specific files from the web application root of its self-hosted products. Atlassian rates it 9.3 (Critical) under CVSS 4.0, with network attack vector, low complexity and no privileges or user interaction. The weakness is CWE-552.

Affected: Bitbucket Data Center, Confluence Data Center, Jira Software Data Center, Jira Service Management Data Center, Bamboo Data Center, Crowd Data Center, Crucible and Fisheye. Atlassian's advisory says all versions before the fixed releases are affected; its CVE record gives the introduction points shown below. Atlassian Cloud has already been patched and needs no customer action. The advisory covers Data Center; it does not list Server editions.

ProductIntroduced in (per Atlassian's CVE record)Fixed versions
Bitbucket Data Center4.6.09.4.26, 10.2.8, 10.5.1
Confluence Data Center5.10.09.2.26, 10.2.19
Jira Software Data Center7.1.09.12.40, 10.3.26, 11.3.12
Jira Service Management Data Center3.1.05.12.40, 10.3.26, 11.3.12
Bamboo Data Center7.0.110.2.24, 12.1.12
Crowd Data Center2.11.06.3.7, 7.0.3, 7.1.7, 7.2.4
Cruciblenot stated4.9.15
Fisheyenot stated4.9.15

Data graphic: table of fixed versions for CVE-2026-21589 across eight Atlassian products, e.g. Bitbucket Data Center 9.4.26, 10.2.8 or 10.5.1; Confluence Data Center 9.2.26 or 10.2.19; Crowd Data Center 6.3.7, 7.0.3, 7.1.7 or 7.2.4; Crucible and Fisheye 4.9.15.

Fixed releases by product; upgrade to the version for your release line or later.

Why it matters

The affected list is Atlassian's whole self-hosted stack: source code, wiki, tickets, builds and identity. (This is a different bug from the earlier MCP Atlassian server flaw.) Atlassian says exploitation needs the exact file name and path and that the bug cannot list directories. That limit removes blind browsing. It does not remove targeting. In our reading, because Atlassian documents where its products keep configuration files, an attacker who knows where a product keeps a sensitive file does not need to enumerate anything. Atlassian itself warns that "in some configurations, there may be sensitive files present that increase your risk." (We covered a similar GitLab arbitrary file read earlier.)

watchTowr Labs, which published its own analysis on 6 October, says the read is confined to the application server's web root (it cannot leave the Tomcat context) but includes the WEB-INF directory, where configuration lives. Its worked example is a Jira instance integrated with Crowd, where a config file holds the Crowd application password in plaintext; the researchers show that credential leading to Crowd administration and then Jira administrator. They note that a Crowd IP allow-list would make that chain harder. That chain is the researchers' demonstrated scenario; see the 16 October update above for the exploitation attempts since reported.

On exposure, watchTowr says a quick internet search finds just under 700,000 Confluence instances and estimates the total across products in the "six to seven-figure range." Those are the researchers' claims with no method given, and they count internet-facing instances, not confirmed vulnerable ones. Atlassian's 9.3 is the only published score; watchTowr gives none, though it writes that it sometimes regrets that CVSS stops at 10 for a flaw baked into almost all of a vendor's products.

Technical details

Atlassian describes it as arbitrary file access within the web application root: a request for a specific file path returns that file without authentication. watchTowr examined Bitbucket, Confluence and Jira and concluded the products share the same underlying behaviour. We do not reproduce request details here; see the watchTowr write-up in the sources.

As of 6 October CISA's Vulnrichment record lists SSVC values of exploitation none, automatable no and technical impact partial. At that point Atlassian said it found no evidence of exploitation of its Cloud products and said nothing about Data Center, and SecurityWeek reported that watchTowr had seen none in the wild. That changed on 7 October: Previdian reported exploitation attempts on its honeypots and watchTowr's FAQ says it has observed in-the-wild exploitation (see the update above). CISA added the CVE to its KEV catalogue on 15 October (see the update above).

What defenders should do

  • Patch to the fixed version for your release line (table above). Atlassian recommends the fixed LTS release or later.
  • If you cannot patch now, take instances off the public internet. Atlassian says this applies even to instances that require user login.
  • Apply one of Atlassian's temporary mitigations: a WAF or proxy regex rule for all products; a Tomcat RewriteValve rule for Confluence, JSM, Jira, Bamboo and Crowd; or a urlrewrite.xml edit for Bitbucket. Atlassian stresses these are interim.
  • Hunt in access logs. Atlassian says to URL-decode each request line (up to two passes) and look for two dots adjacent to a slash, backslash or double colon, or to search raw lines with its regex.
  • Treat credentials in files under WEB-INF as exposed on any instance that was reachable and unpatched, and rotate them, starting with the Crowd application password in crowd.properties. Then check Crowd and Jira for administrator accounts you did not create.
  • Restrict which addresses can reach Crowd (an allow-list). watchTowr says this makes the credential chain significantly harder.
  • US federal agencies must patch by 18 October and carry out CISA's BOD 26-04 forensic triage, or stop using the product if they cannot mitigate (CISA KEV required action).
  • Atlassian cannot confirm whether your instance was affected; involve your security team.

What is still unclear

  • Whether Crucible and Fisheye share the introduction points above; Atlassian's CVE record gives none for them.
  • How many real instances were hit. Previdian counts 607 honeypot attempts from 65 IP addresses and VulnCheck cites Bamboo targeting (which products were actually hit is unclear), but no source has published victim counts or confirmed compromises, and Atlassian's no-evidence statement covers Cloud only.
  • The real number of exposed, unpatched instances. The 700,000 figure is a researcher claim.
  • Who found it. Atlassian lists no credits.

Sources

Keep reading

All latest →
  1. criticalExploitsExploited FreeSWITCH Flaw Turns One SIP MESSAGE Into Command Execution: Update to 1.11.34 min
  2. highExploitsFortiManager FGFM Flaw Lets Unauthenticated Attackers Run Code5 min
  3. highExploitsAnthropic's Scanner and Google Both Found a Critical Apache httpd mod_http2 RCE: Update to 2.4.714 min
  4. highExploitsMalbear Says an AI Agent Breached a Telco via Cisco BroadWorks CVE-2026-20320; CISA Lists It, Cisco Sees No Evidence5 min
  5. criticalExploitsCheck Point Gaia Root Flaw Is Exploited, and Attackers Chained It With PHP2Shell to Wipe ESXi Hosts4 min
  6. highExploitsNovember 2026 Patch Tuesday: Patch the Exploited Outlook Hash Leak First, Then Two Zero-Click 9.8 RCEs4 min