Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-108855 Overview
CVE-2026-108855 is a missing authorization vulnerability in UnicomAI Wanwu through version 0.6.5. The flaw resides in the application's unpublish endpoint, which fails to verify that the requesting user owns the targeted AppKey. Any authenticated enabled user can supply a target appId and appType from the exploration marketplace to delete other users' api_key rows across organizations. Successful exploitation cuts off Model Context Protocol (MCP) and OpenAPI client access for the targeted consumers. The vulnerability is classified under CWE-862: Missing Authorization.
Critical Impact
Authenticated users can revoke arbitrary AppKeys belonging to other tenants, disrupting MCP and OpenAPI integrations across organizational boundaries.
Affected Products
- UnicomAI Wanwu through version 0.6.5
- Wanwu app-service component handling publish/unpublish operations
- Downstream MCP and OpenAPI clients that depend on issued AppKeys
Discovery Timeline
- 2026-10-11 - CVE-2026-108855 published to NVD
- 2026-10-16 - Last updated in NVD database
Technical Details for CVE-2026-108855
Vulnerability Analysis
UnicomAI Wanwu is an open-source platform for building and publishing AI applications. The platform issues AppKeys that downstream MCP and OpenAPI clients use to authenticate against published apps. The unpublish endpoint DELETE /v1/appspace/app/publish is intended to let app owners revoke their own publication state and associated API keys.
The handler trusts the appId and appType parameters supplied by the caller without checking whether the authenticated user owns the target app or its associated AppKey. Any enabled user who can browse the exploration marketplace can enumerate valid appId and appType values and submit them to the endpoint. The backend then deletes the matching api_key rows from the database, severing active integrations.
The scope spans tenants because the authorization check is absent entirely, not merely scoped incorrectly. See the VulnCheck advisory and the HackMD technical writeup for additional context.
Root Cause
The root cause is a missing ownership check in the ORM query path. The affected logic lives in internal/app-service/client/orm/app.go lines 48-68. The query deletes api_key rows based solely on appId and appType without constraining by the authenticated user's userId or orgId.
Attack Vector
An attacker authenticates with any enabled Wanwu account. They browse the exploration marketplace to identify a victim app's appId and appType. They then issue a DELETE /v1/appspace/app/publish request supplying those identifiers. The backend removes the victim's AppKey records, breaking any MCP or OpenAPI consumer that relied on them. No code execution is required and no data is read; the impact is integrity and availability loss for API-key-dependent integrations.
No proof-of-concept code is published for this issue. Refer to the VulnCheck advisory for request-level detail.
Detect Identity Attacks the Moment Authentication Fails
Singularity™ Identity surfaces credential misuse, lateral movement, and privilege escalation in real time, so you can contain the breach before it spreads.
Stop Identity AttacksDetection Methods for CVE-2026-108855
Indicators of Compromise
- Unexpected DELETE /v1/appspace/app/publish requests originating from accounts that do not own the referenced appId.
- Sudden removal of rows from the api_key table without a corresponding owner-initiated unpublish action in audit logs.
- Downstream MCP or OpenAPI clients returning 401/403 errors immediately after an unpublish call from an unrelated tenant.
Detection Strategies
- Correlate application access logs for DELETE /v1/appspace/app/publish with the ownerId of the targeted appId and alert on mismatches.
- Instrument database triggers or change-data-capture on the api_key table to flag deletions that cross tenant boundaries.
- Review marketplace browsing patterns followed shortly by unpublish calls as a behavioral signal of enumeration-then-abuse.
Monitoring Recommendations
- Centralize Wanwu app-service logs in a SIEM and build a rule for cross-tenant unpublish attempts.
- Baseline normal AppKey lifecycle events per tenant and alert on statistical anomalies in deletion rates.
- Monitor downstream MCP/OpenAPI client error telemetry for sudden spikes in authentication failures.
How to Mitigate CVE-2026-108855
Immediate Actions Required
- Restrict access to the Wanwu app-service to trusted users until a fixed version is deployed.
- Audit the api_key table for unexpected deletions since deployment and reissue AppKeys to affected consumers.
- Review recent DELETE /v1/appspace/app/publish activity and disable any accounts observed targeting apps they do not own.
Patch Information
No official fixed version is listed in the NVD entry for releases beyond 0.6.5. Track the UnicomAI Wanwu GitHub repository for updated releases and apply any security patch that enforces an ownership check on the unpublish handler and underlying ORM query.
Workarounds
- Place the unpublish endpoint behind a reverse proxy or API gateway rule that validates the authenticated user owns the submitted appId before forwarding the request.
- Reduce the pool of enabled users with marketplace access to the minimum required set, lowering the attacker population.
- Add a database-level policy or application-layer middleware that scopes api_key deletions by userId and orgId as a defense-in-depth control until an upstream fix is available.
# Example reverse-proxy guard (nginx) - block unpublish calls lacking owner context
location = /v1/appspace/app/publish {
limit_except GET POST { deny all; }
# Require an internal header set by an auth sidecar that verifies ownership
if ($http_x_wanwu_owner_verified != "1") { return 403; }
proxy_pass http://wanwu-app-service;
}
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

