What happened
Apache disclosed CVE-2026-90436 in Apache Axis 1.x on November 6, 2026. Rapid7 subsequently tied the end-of-life SOAP component to exposed Confluence Data Center and N-able N-central deployments. On November 18, 2026, CISA added CVE-2026-90436 to KEV with a federal remediation due date of November 21, 2026, converting an older Apache Axis disclosure into a compromise-assessment priority. CVE-2026-90436 is an unsafe reflection flaw in end-of-life Apache Axis 1.x through 1.4; Apache does not plan a fixed Axis 1.x release.
Rapid7 observed exploitation on November 13, 2026 against Confluence Data Center with Remote API (XML-RPC & SOAP) enabled. Rapid7 observed unauthenticated POST requests to /plugins/servlet/soap-axis1/confluenceservice-v2 returning HTTP 500 while the Confluence Java process launched Bash and curl. Just over three minutes after initial execution, the intruder used Confluence database credentials against MySQL; roughly four hours later a service credential reached a production cloud data warehouse; approximately twelve hours later an integration account signed in to Microsoft 365 from 192.238.253[.]176.
The attacker ran hundreds of hex-encoded commands, used temporary Pinggy and Cloudflare Tunnel infrastructure, searched /var/atlassian/application-data/confluence/attachments/v4/, and invoked Impacket GetNPUsers.py in an unsuccessful AS-REP roasting attempt. The published Confluence indicators are 192.238.253[.]176, 5.133.47[.]243, 163.182.68[.]109, 104.13.106[.]166, *.trycloudflare[.]com, wpefh-192-238-253-176.free.pinggy[.]net and webhook[.]site; N-able additionally published 157.20.241[.]16, 139.99.69[.]109 and 79.76.51[.]56 for N-central environments with Legacy API exposure.
Confluence Data Center 9.2.28 is fixed; version 10.x removed the vulnerable Remote API feature, and Atlassian Cloud is unaffected. Atlassian says the mitigation is validated when the endpoint returns HTTP Status 403 - Forbidden. N-central 2026.4 HF3 build 2026.4.1.39 permanently blocks legacy APIs; hosted and on-premises customers must also upgrade Agents and Probes. Attribution posture: Rapid7 did not attribute the observed exploitation to a named actor, and CISA’s KEV entry does not identify one.
Why this matters now
CISA’s KEV addition changes the operating assumption. This is no longer a theoretical dependency flaw or a routine Confluence patch: active exploitation is authoritative, federal guidance requires forensic triage, and private-sector organisations should apply the same distinction between remediation and compromise assessment. Rapid7’s observed intrusion progressed from a crafted SOAP request to database use, a production cloud data warehouse and a Microsoft 365 integration identity.
The exposure is concentrated in privileged platforms. Confluence can hold credentials, architecture records, attachments and integration secrets; N-central can administer downstream endpoints. An attacker reaching either platform may inherit trusted access paths that substantially exceed the initial server. The end-of-life Axis dependency also means upstream package replacement is not a normal upgrade path, forcing product-specific remediation and a wider search for other embedded copies.
Not every deployment is vulnerable: Confluence requires Remote API (XML-RPC & SOAP), while N-central exposure centres on Legacy API. That conditionality should accelerate configuration verification, not reduce urgency. Internet-facing systems with those features enabled require immediate isolation or remediation, and internal systems still require review because compromised credentials or adjacent access can make them reachable.
The decision for security leaders
Separate the remediation workstream from incident response. Product owners can patch or disable the affected APIs, but the SOC and incident-response lead must independently determine whether exploitation occurred while the interface was reachable. Closure requires both results; a clean version scan alone does not answer whether credentials or connected systems were reached.
Make application owners responsible for dependency reachability, not merely package presence. Axis 1.x is embedded inside supported enterprise products and may also remain in custom Java services. Require a decision-grade inventory that records the product, version, API state, network exposure, business owner and evidence source for every potentially reachable deployment.
Treat confirmed server execution as an identity-containment event. The observed path reached stored database credentials and an integration identity, so responders should scope service accounts, API credentials, cloud sessions and secrets accessible from the affected platform. Involve data owners if Confluence attachments, connected warehouses or managed customer endpoints were reachable.
Evidence of closure
- A signed inventory identifies every Axis-bearing product, its owner, network exposure and API state.
- Each Confluence instance records version 9.2.28 or current 10.x, or returns HTTP Status 403 at the SOAP endpoint.
- Each N-central server records build 2026.4.1.39 with upgraded Agents and Probes.
- SOC results document the disposition of every IOC, Axis request and Java child-process match examined in scope_reviewed_2026_11_19.pdf .
The Security.io assessment
The evidence supports a high-confidence active-exploitation assessment for CVE-2026-90436. CISA’s KEV decision is authoritative, Atlassian defines the exact Confluence exposure condition, N-able has published relevant indicators and a hardened release, and Rapid7 provides detailed endpoint and application telemetry from an observed intrusion.
The evidence does not establish that every Axis deployment, Confluence server or N-central instance has been targeted. Confluence requires a feature that is disabled by default on newer installations, and the public N-central linkage includes product-specific conditions. Those limitations narrow the exposure cohort but do not reduce the need for decisive verification on privileged systems.
Indicator absence is not sufficient closure. Temporary tunnel domains can rotate, exploitation traffic may return HTTP 500, and the strongest evidence may be Java child processes, unexpected credential use or access to connected data services. The correct conclusion is therefore conditional: patched and unexposed systems may close as vulnerability work; systems exposed during the attack window require documented compromise assessment.
Questions for the morning meeting
- Can the organisation identify every product and custom service that embeds Apache Axis 1.x?
- Which Confluence and N-central systems exposed the vulnerable APIs before remediation?
- Can owners prove compromise assessment separately from patch or configuration status?
- Which integrations prevent immediate retirement of the affected SOAP interfaces?