Axis RCE and Confluence Credential Harvesting
Critical severity TLP:CLEAR Part 1 of 2 Research by Rapid7 Published 18 Nov 2026
Hypothesis
An adversary exploits the Apache Axis SOAP vulnerability (CVE-2026-90436) in Atlassian Confluence to execute shell commands and harvest credentials from the attachment store.
Brief
Why this huntaAdversaries currently exploit a critical remote code execution vulnerability in
Apache Axis (CVE-2026-90436) to target Atlassian Confluence Data Center. Rapid7 reported this activity in their Rapid7 (https://www.rapid7.com/blog/post/etr-rapid7-detected-exploitation-of-cve-2026-90436-against-atlassian-confluence-data-center-n-able-n-central-also-exploited) analysis, noting that attackers use the vulnerability to gain initial access and then move to harvest sensitive data. This hunt provides a structured approach to identifying both the initial SOAP exploit and the subsequent post-exploitation behaviors that signify a successful breach.aa
The
HypothesisaAn adversary exploits the Apache Axis SOAP vulnerability in Confluence to execute shell commands. Once they establish a foothold, they use automated scripts to discover and harvest credentials or sensitive documents from the Confluence attachment store, often exfiltrating this data through public tunneling or webhook services.aa
How the Hunt
FlowsaThe hunt uses a gated execution model to reduce false positives and focus analyst attention on high-confidence leads. The process starts at the HTTP surface, searching for specific POST requests to the Axis SOAP endpoint that resulted in an Internal Server Error (500). This behavior matches the observed exploitation pattern where the server fails to process the malicious SOAP payload but executes the embedded commands.aaIf the scoping query identifies suspicious hosts, the hunt fans out into three parallel investigations. The first check monitors the process surface for Java spawning bash, sh, or curl. Because the Confluence service account rarely executes shell utilities directly from the Java process, this signal provides strong evidence of remote code execution. The second phase examines file activity for rare access to the Confluence attachment directory. Attackers often target the /attachments/v4/ path to find sensitive corporate data. By baselining normal file access across the fleet, the hunt highlights scripts or rare processes that interact with these files.aaThe final corroboration phase looks for DNS resolutions of tunneling services like Pinggy and Cloudflare Tunnels. Adversaries use these services to bypass firewall restrictions and establish command-and-control channels without requiring incoming connections. An analyst then triages the collected evidence from all four surfaces—HTTP, process, file, and DNS—to reach a final verdict. If the results confirm an intrusion, the hunt playbook provides instructions for server isolation and forensic collection.aa
What this
Hunt Cannot SeeaThe effectiveness of the initial lead depends on the retention of HTTP access logs with response codes. If the environment lacks this telemetry, the hunt may fail to identify the entry point. Additionally, if an adversary uses a custom, low-prevalence tunneling tool not included in the provided parameters, the DNS query will not flag the exfiltration. The hunt also assumes the adversary uses the default Confluence attachment path; custom configurations may require manual adjustment of the file search parameters.aa
How to
Run itaThis hunt is packaged as an open hunt.md playbook. You can import it into Huntbase or any hunt.md-aware runtime. This format allows you to run the queries directly against your telemetry and manage the multi-phase flow through a single interface.
In this series
Steps
-
Query, scoping
Lead: SOAP exploit requests
Identify potential Axis exploitation by finding HTTP POST requests to the SOAP endpoint that resulted in an Internal Server Error (500), matching the behavior observed in successful compromises.
sql, reads hb_http_activitySELECT device_hostname, src_endpoint_ip, url_path, status_code, timeFROM hb_http_activityWHERE instr(',' || '{{vulnerable_uris}}' || ',', ',' || LOWER(url_path) || ',') > 0 AND status_code = 500 AND http_method = 'POST' AND time >= datetime('now', '-{{lookback_days}} days')What a hit looks like. Rows naming hosts that received suspicious POST requests to the Axis endpoint. Silence suggests no active targeting of the SOAP API in the window.
-
Agent triage
Evaluate lead
Assess whether the HTTP 500 errors found in the lead query represent high-confidence exploit attempts against Confluence.
-
Decision
Gate on lead
Halt the hunt if no suspicious lead exists, or proceed to expensive behavioral queries if exploit traffic is confirmed.
-
Query, detection candidate
Java shell execution
Identify Confluence's Java process spawning bash or network utilities, which indicates successful remote code execution from the Axis exploit.
sql, reads hb_process_activitySELECT device_hostname, process_name, process_cmd_line, parent_process_name, user_name, timeFROM hb_process_activityWHERE ('{{scope_hosts}}' = '' OR instr(',' || '{{scope_hosts}}' || ',', ',' || device_hostname || ',') > 0) AND LOWER(parent_process_name) LIKE '%java%' AND (LOWER(process_name) LIKE '%bash%' OR LOWER(process_name) LIKE '%sh%' OR LOWER(process_name) LIKE '%curl%') AND time >= datetime('now', '-{{lookback_days}} days')What a hit looks like. A shell or curl process running under the confluence service account with a Java parent. This is a high-confidence indicator of exploitation.
-
Query, baseline
Attachment access prevalence
Identify rare processes accessing the Confluence attachment directory, revealing the automated data harvesting observed in the report.
sql, reads hb_file_activitySELECT device_hostname, LOWER(process_name) AS proc, COUNT(DISTINCT device_hostname) AS hosts, MIN(time) AS first_seenFROM hb_file_activityWHERE file_path LIKE '%/attachments/v4/%' AND time >= datetime('now', '-{{lookback_days}} days')GROUP BY procHAVING hosts <= 3What a hit looks like. Rare processes interacting with the attachment store. Normal web server activity should appear across the fleet; attacker scripts will be rare.
-
Query, enrichment
Tunnel DNS activity
Look for DNS resolutions of Pinggy, Cloudflare Tunnels, and Webhook.site, which attackers use to receive shell commands and exfiltrate harvested data.
sql, reads hb_dns_activitySELECT device_hostname, query_hostname, process_name, COUNT(*) AS lookupsFROM hb_dns_activityWHERE (instr(',' || '{{tunnel_domains}}' || ',', ',' || LOWER(query_hostname) || ',') > 0 OR LOWER(query_hostname) LIKE '%.trycloudflare.com' OR LOWER(query_hostname) LIKE '%.pinggy.net') AND time >= datetime('now', '-{{lookback_days}} days')GROUP BY device_hostname, query_hostname, process_nameWhat a hit looks like. Hosts resolving tunnel provider subdomains. This corroborates C2 activity following a successful exploit.
-
Agent triage
Triage intrusion
Correlate the exploit lead with the observed shell execution and data harvesting evidence to reach a final verdict per host.
-
Decision
Route on triage
Initiate response actions for compromised hosts or route suspicious findings for analyst review.
-
Response action
Isolate server
Prevent further exfiltration of harvested credentials by isolating the compromised server.
-
Analyst task
Analyst review
Manually verify the intrusion and prepare for incident response if any credential harvesting is confirmed.
-
Analyst task
Close out
Finalize the hunt and record negative results.
Coverage
3 of 5 stages covered
| Stage and ATT&CK technique | Covered | How, or why not |
|---|---|---|
| Unauthenticated Axis SOAP RCE T1190 |
Yes | soap-exploit-leads |
| Automated shell execution and tunneling T1059.004 T1105 T1572 |
Yes | java-shell-execution, tunnel-dns-activity |
| Confluence data and credential harvesting T1083 T1552.001 T1213 |
Yes | attachment-access-prevalence |
| Identity service enumeration and AS-REP roasting T1087.002 T1558.004 |
Out of scope | Belongs to another part of the 'Rapid7 Detected Exploitation of CVE-2026-90436 against Atlassian Confluence Data Center, N-able N-central also Exploited' series. |
| Cloud data warehouse and M365 access T1537 T1078.004 |
Out of scope | Belongs to another part of the 'Rapid7 Detected Exploitation of CVE-2026-90436 against Atlassian Confluence Data Center, N-able N-central also Exploited' series. |
Blind spots
- Needs hb_http_activity with response status code. Missing the primary indicator of the initial exploit, leading to a false negative for the gated hunt. It would answer: whether the specific vulnerable Axis endpoint was targeted and returned a 500 error
- Needs 30 days of hb_process_activity retention. An intrusion that occurred several weeks ago might be missed if process logs rotate too quickly. It would answer: whether the automated shell activity occurred outside the current telemetry window
- Needs an endpoint agent on every server. An unmanaged Confluence instance provides an invisible foothold for the attacker. It would answer: whether an unmonitored server is currently compromised
Parameters and data
Parameters
| Parameter | Type | Default | What it is |
|---|---|---|---|
lookback_days | number | 14 | Days of history to examine. |
scope_hosts | list[host] | none | Hosts identified as potential exploit targets from the lead query; paste these back in to refine the fan-out queries. |
tunnel_domains | list[domain] | webhook.site | Base domains for tunneling and exfiltration services identified in the report. |
vulnerable_uris | list[string] | /plugins/servlet/soap-axis1/confluenceservice-v2 | The vulnerable Apache Axis SOAP endpoint in Confluence. |
Data sources
| Source | Category | Telemetry |
|---|---|---|
| Endpoint telemetry (hb_ surfaces) | endpoint | endpoint |
| Web server / proxy logs | siem | network |
Source
An open hunt.md file. It runs anywhere that reads the format.
---
analysis: A single detection rule may catch the Java-to-bash execution, but this hunt
correlates the initial HTTP exploit attempt, the use of tunneling services, and
rare access to the attachment store, providing the full context of a credential
harvesting campaign.
blind_spots:
- id: no-access-log-detail
question: whether the specific vulnerable Axis endpoint was targeted and returned
a 500 error
requires: hb_http_activity with response status code
risk: Missing the primary indicator of the initial exploit, leading to a false negative
for the gated hunt.
stage: initial-access-axis-soap-rce
- id: limited-retention
question: whether the automated shell activity occurred outside the current telemetry
window
requires: 30 days of hb_process_activity retention
risk: An intrusion that occurred several weeks ago might be missed if process logs
rotate too quickly.
stage: execution-automated-shell-c2
- id: no-agent-coverage
question: whether an unmonitored server is currently compromised
requires: an endpoint agent on every server
risk: An unmanaged Confluence instance provides an invisible foothold for the attacker.
coverage:
- stage: initial-access-axis-soap-rce
status: covered
steps:
- soap-exploit-leads
- stage: execution-automated-shell-c2
status: covered
steps:
- java-shell-execution
- tunnel-dns-activity
- stage: discovery-credential-harvesting
status: covered
steps:
- attachment-access-prevalence
- reason: Belongs to another part of the 'Rapid7 Detected Exploitation of CVE-2026-90436
against Atlassian Confluence Data Center, N-able N-central also Exploited' series.
stage: lateral-movement-identity-pivot
status: out_of_scope
- reason: Belongs to another part of the 'Rapid7 Detected Exploitation of CVE-2026-90436
against Atlassian Confluence Data Center, N-able N-central also Exploited' series.
stage: exfiltration-cloud-warehouse-access
status: out_of_scope
guardrails:
claims: no_unsupported
evidence: citation_required
missing_data: not_benign
telemetry: untrusted
hunt:
applicability: campaign-specific
handoff: promote-to-detection
justification: Exploitation of CVE-2026-90436 allows unauthenticated RCE on Confluence
servers, making it a critical threat to intellectual property and internal credentials.
Identifying post-exploit harvesting behavior provides high assurance that the
organization is not compromised.
methodology: model-assisted
trigger: intel-report
hypothesis: An adversary exploits the Apache Axis SOAP vulnerability (CVE-2026-90436)
in Atlassian Confluence to execute shell commands and harvest credentials from the
attachment store.
labels:
- hunt
- attack.t1190
- attack.t1059.004
- attack.t1105
- attack.t1572
- attack.t1083
- attack.t1552.001
- attack.t1213
- discovery
- execution
- exfiltration
- initial access
- lateral movement
name: Axis RCE and Confluence Credential Harvesting
parameters:
lookback_days:
default: '14'
description: Days of history to examine.
type: number
scope_hosts:
default: []
description: Hosts identified as potential exploit targets from the lead query;
paste these back in to refine the fan-out queries.
type: list[host]
tunnel_domains:
default:
- webhook.site
description: Base domains for tunneling and exfiltration services identified in
the report.
from:
kind: article
observed: '2026-11-16'
ref: Rapid7
type: list[domain]
vulnerable_uris:
default:
- /plugins/servlet/soap-axis1/confluenceservice-v2
description: The vulnerable Apache Axis SOAP endpoint in Confluence.
from:
kind: article
observed: '2026-11-16'
ref: Rapid7
type: list[string]
provenance:
authors:
- name: Huntbase hunt generation
org: huntbase.io
generated:
by: huntbase-hunt-generation
from: https://www.rapid7.com/blog/post/etr-rapid7-detected-exploitation-of-cve-2026-90436-against-atlassian-confluence-data-center-n-able-n-central-also-exploited
gates:
- dry-run
- lint
- critic
model: hb_google/gemini-3-flash-preview
rationale: Focus on internet-exposed Atlassian Confluence Data Center servers. The
lead query is highly effective if HTTP access logs are retained; if not, proceed
directly to the process execution step using the whole estate.
references:
- name: "Rapid7 \u2014 ETR: Rapid7 Detected Exploitation of CVE-2026-90436 against\
\ Atlassian Confluence Data Center"
url: https://www.rapid7.com/blog/post/etr-rapid7-detected-exploitation-of-cve-2026-90436-against-atlassian-confluence-data-center-n-able-n-central-also-exploited
related:
- hunt: lateral-movement-identity-pivot
reason: This hunt focuses on the initial Confluence foothold and local harvesting;
the subsequent identity pivot to cloud resources belongs in a follow-on identity
hunt.
relation: out-of-scope-alternative
scenario:
stages:
- name: Unauthenticated Axis SOAP RCE
observables:
- POST /plugins/servlet/soap-axis1/confluenceservice-v2
- HTTP 500 Internal Server Error
- CVE-2026-90436
slug: initial-access-axis-soap-rce
tactic: initial-access
techniques:
- T1190
- name: Automated shell execution and tunneling
observables:
- confluence service account
- Java process launching bash and curl
- hex-encoded shell commands
- trycloudflare.com
- free.pinggy.net
- webhook.site
- wpefh-192-238-253-176.free.pinggy.net
- sequential task IDs
- labeled delimiters in command output
slug: execution-automated-shell-c2
tactic: execution
techniques:
- T1059.004
- T1105
- T1572
- name: Confluence data and credential harvesting
observables:
- /var/atlassian/application-data/confluence/attachments/v4/
- MySQL authentication using Confluence credentials
- parsing Office documents for credentials
- Luhn checks for payment card numbers
- searching for 'VPN', 'token', 'encryption keys'
slug: discovery-credential-harvesting
tactic: discovery
techniques:
- T1083
- T1552.001
- T1213
- name: Identity service enumeration and AS-REP roasting
observables:
- Atlassian Crowd API queries
- Impacket GetNPUsers.py against Domain Controller
- AS-REP roasting
slug: lateral-movement-identity-pivot
tactic: lateral-movement
techniques:
- T1087.002
- T1558.004
- name: Cloud data warehouse and M365 access
observables:
- Production cloud data warehouse queries
- M365 sign-in via legacy authentication
- IP 192.238.253.176
slug: exfiltration-cloud-warehouse-access
tactic: exfiltration
techniques:
- T1537
- T1078.004
summary: Attackers exploit a critical unauthenticated RCE in Apache Axis (CVE-2026-90436)
affecting Atlassian Confluence and N-able N-central to gain initial footprints.
Using AI-assisted automation, they harvest credentials from local databases and
file attachments before pivoting to internal identity services and production
cloud environments.
series:
index: 1
slug: rapid7-detected-exploitation-of-cve-2026-90436-against-atlassian-confluence-data-center-n-able-n
title: Rapid7 Detected Exploitation of CVE-2026-90436 against Atlassian Confluence
Data Center, N-able N-central also Exploited
total: 2
severity: critical
targets:
analyst:
name: Tier-2 analyst
role: analyst
endpoint:
category: endpoint
name: Endpoint telemetry (hb_ surfaces)
telemetry:
- endpoint
hunter:
agent: true
name: Hunt agent
web:
category: siem
name: Web server / proxy logs
telemetry:
- network
tlp: clear
type: investigation
---
# Axis RCE and Confluence Credential Harvesting
The adversary exploits a critical remote code execution vulnerability in Apache Axis. This hunt identifies the attack using a gated flow. The hunt first searches for HTTP 500 errors on the Axis SOAP endpoint. If potential exploit traffic is found, the hunt fans out to identify anomalous child processes spawned by the Java application, rare access to the Confluence attachment directory, and connections to known tunneling services. This approach allows an analyst to confirm the initial exploit, the execution of automated shells, and the subsequent harvesting of sensitive documents.
## soap-exploit-leads
<!-- Lead: SOAP exploit requests -->
Identify potential Axis exploitation by finding HTTP POST requests to the SOAP endpoint that resulted in an Internal Server Error (500), matching the behavior observed in successful compromises.
```sqlite target=web role=scoping params=(lookback_days=lookback_days, vulnerable_uris=vulnerable_uris)
~~~yaml
expected: Rows naming hosts that received suspicious POST requests to the Axis endpoint.
Silence suggests no active targeting of the SOAP API in the window.
reads:
- device_hostname
- src_endpoint_ip
- url_path
- status_code
- time
- http_method
silence: not_evidence_of_absence
source: hb_http_activity
verified: dry-run
verified_at: '2026-11-18'
~~~
SELECT device_hostname, src_endpoint_ip, url_path, status_code, time FROM hb_http_activity WHERE instr(',' || '{{vulnerable_uris}}' || ',', ',' || LOWER(url_path) || ',') > 0 AND status_code = 500 AND http_method = 'POST' AND time >= datetime('now', '-{{lookback_days}} days')
```
## evaluate-lead
<!-- Evaluate lead -->
```agent target=hunter
cite: required
context:
- soap-exploit-leads
max_iterations: 3
objective: Determine if the observed SOAP requests indicate a confirmed exploit attempt
by analyzing source IP and request patterns.
success_criteria: A verdict of malicious or suspicious for at least one host, citing
specific rows.
tools:
- endpoint
- web
```
## gate-on-lead
<!-- Gate on lead -->
if~: "the evaluate-lead verdict identifies at least one host with suspicious SOAP exploit attempts" (confidence: high, judge=hunter)
then: → corroborate-activity
indeterminate: → analyst-review
unavailable: → analyst-review (blind_spot: no-access-log-detail)
else: → close-out
## corroborate-activity
<!-- Corroborate activity -->
parallel:
- → java-shell-execution
- → attachment-access-prevalence
- → tunnel-dns-activity
join: → triage-intrusion
## java-shell-execution
<!-- Java shell execution -->
Identify Confluence's Java process spawning bash or network utilities, which indicates successful remote code execution from the Axis exploit.
```sqlite target=endpoint role=detection-candidate params=(lookback_days=lookback_days, scope_hosts=scope_hosts)
~~~yaml
expected: A shell or curl process running under the confluence service account with
a Java parent. This is a high-confidence indicator of exploitation.
reads:
- device_hostname
- process_name
- process_cmd_line
- parent_process_name
- user_name
- time
silence: not_evidence_of_absence
source: hb_process_activity
verified: dry-run
verified_at: '2026-11-18'
~~~
SELECT device_hostname, process_name, process_cmd_line, parent_process_name, user_name, time FROM hb_process_activity WHERE ('{{scope_hosts}}' = '' OR instr(',' || '{{scope_hosts}}' || ',', ',' || device_hostname || ',') > 0) AND LOWER(parent_process_name) LIKE '%java%' AND (LOWER(process_name) LIKE '%bash%' OR LOWER(process_name) LIKE '%sh%' OR LOWER(process_name) LIKE '%curl%') AND time >= datetime('now', '-{{lookback_days}} days')
```
## attachment-access-prevalence
<!-- Attachment access prevalence -->
Identify rare processes accessing the Confluence attachment directory, revealing the automated data harvesting observed in the report.
```sqlite target=endpoint role=baseline params=(lookback_days=lookback_days)
~~~yaml
baseline:
compare: first_seen
window: '{{lookback_days}}d'
expected: Rare processes interacting with the attachment store. Normal web server
activity should appear across the fleet; attacker scripts will be rare.
prevalence:
by: device_hostname
key:
- proc
rare_below: 3
reads:
- device_hostname
- process_name
- file_path
- time
silence: not_evidence_of_absence
source: hb_file_activity
verified: dry-run
verified_at: '2026-11-18'
~~~
SELECT device_hostname, LOWER(process_name) AS proc, COUNT(DISTINCT device_hostname) AS hosts, MIN(time) AS first_seen FROM hb_file_activity WHERE file_path LIKE '%/attachments/v4/%' AND time >= datetime('now', '-{{lookback_days}} days') GROUP BY proc HAVING hosts <= 3
```
## tunnel-dns-activity
<!-- Tunnel DNS activity -->
Look for DNS resolutions of Pinggy, Cloudflare Tunnels, and Webhook.site, which attackers use to receive shell commands and exfiltrate harvested data.
```sqlite target=endpoint role=enrichment params=(lookback_days=lookback_days, tunnel_domains=tunnel_domains)
~~~yaml
expected: Hosts resolving tunnel provider subdomains. This corroborates C2 activity
following a successful exploit.
reads:
- device_hostname
- query_hostname
- process_name
- time
silence: not_evidence_of_absence
source: hb_dns_activity
verified: dry-run
verified_at: '2026-11-18'
~~~
SELECT device_hostname, query_hostname, process_name, COUNT(*) AS lookups FROM hb_dns_activity WHERE (instr(',' || '{{tunnel_domains}}' || ',', ',' || LOWER(query_hostname) || ',') > 0 OR LOWER(query_hostname) LIKE '%.trycloudflare.com' OR LOWER(query_hostname) LIKE '%.pinggy.net') AND time >= datetime('now', '-{{lookback_days}} days') GROUP BY device_hostname, query_hostname, process_name
```
## triage-intrusion
<!-- Triage intrusion -->
```agent target=hunter
cite: required
context:
- evaluate-lead
- java-shell-execution
- attachment-access-prevalence
- tunnel-dns-activity
max_iterations: 6
objective: Determine if the evidence across HTTP, process, file, and DNS surfaces
confirms a successful breach and data harvesting on the Confluence server.
success_criteria: A final verdict of malicious, suspicious, or benign per host, citing
evidence from all four surfaces.
tools:
- endpoint
- web
```
## route-on-triage
<!-- Route on triage -->
if~: "the triage-intrusion verdict is malicious for at least one host" (confidence: high, judge=hunter)
then: → isolate-server
indeterminate: → analyst-review
unavailable: → analyst-review (blind_spot: limited-retention)
else: → close-out
## isolate-server
<!-- Isolate server -->
```action target=endpoint
~~~yaml
approval: required
~~~
Isolate the host from the network and preserve all logs for forensic analysis. Collect copies of any scripts found in the Confluence temporary directories.
```
→ analyst-review
## analyst-review
<!-- Analyst review -->
```manual target=analyst
Review the cited SOAP requests and shell execution logs. Verify the contents of the attachment directory to identify which specific sensitive documents were accessed by the attacker's scripts.
```
→ end
## close-out
<!-- Close out -->
```manual target=analyst
Record the search parameters and the absence of identified indicators. Re-verify that the Remote XML-RPC & SOAP API is disabled on Confluence instances where possible.
```
→ end
Run this hunt against your own data.
In Huntbase, Scout runs every step across the tools you already connect, checks back through your history and returns a verdict with the evidence. The hunt.md file is open, so you can also take it anywhere that reads the format.
Drafted by machine (huntbase-hunt-generation, hb_google/gemini-3-flash-preview), checked by dry-run, lint, critic, then reviewed by a person.