Skip to main content
Vulnerability Database/CVE-2026-108853

CVE-2026-108853: UnicomAI Wanwu Auth Bypass Vulnerability

CVE-2026-108853 is an authentication bypass flaw in UnicomAI Wanwu that allows low-privileged users to delete other tenants' applications. This post covers the technical details, affected versions, and mitigation steps.

Published:

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-2026-108853 Overview

CVE-2026-108853 is an Insecure Direct Object Reference (IDOR) vulnerability in UnicomAI Wanwu before version 0.6.3. The flaw resides in the DELETE /v1/appspace/app endpoint, which fails to verify tenant ownership before deleting agent or Retrieval-Augmented Generation (RAG) applications. An authenticated low-privileged user can supply another tenant's appId to permanently delete that tenant's applications, workflows, conversations, and associated data. The weakness is classified under CWE-639: Authorization Bypass Through User-Controlled Key.

Critical Impact

Any authenticated tenant can enumerate sequential assistant IDs and issue cross-tenant delete requests, resulting in permanent destruction of other tenants' AI applications and operational data.

Affected Products

  • UnicomAI Wanwu versions prior to 0.6.3
  • The bff-service component handling the appspace API (internal/bff-service/service/appspace.go)
  • Deployments exposing the multi-tenant Wanwu platform to shared authenticated users

Discovery Timeline

  • 2026-10-11 - CVE-2026-108853 published to the National Vulnerability Database (NVD)
  • 2026-10-16 - Last updated in NVD database

Technical Details for CVE-2026-108853

Vulnerability Analysis

UnicomAI Wanwu is a multi-tenant platform for building AI agents and RAG applications. The DELETE /v1/appspace/app handler in internal/bff-service/service/appspace.go accepts an appId from the authenticated caller and routes the delete request to the backing app-service. The handler does not confirm that the supplied appId belongs to the caller's tenant or user. Because assistant identifiers are sequential and guessable, an attacker holding any valid low-privileged account can iterate through IDs to delete arbitrary tenants' resources. Deletion cascades across the application record, its workflows, saved conversations, and derived data, producing permanent data loss with no built-in restore path.

Root Cause

The delete pipeline in the pre-patch client interface accepted only the resource identifier, as shown by signatures such as DeleteApiKey(ctx, keyId) and DelAppKey(ctx, appKeyId). Downstream SQL executed WHERE id = ? without filtering by the caller's userId or orgId. The same pattern existed in the appspace delete flow, leaving ownership verification entirely absent at the service boundary.

Attack Vector

Exploitation requires only network reach to the API and a valid low-privileged session. An attacker enumerates sequential appId values and submits them via DELETE /v1/appspace/app. Each successful request destroys the target tenant's assistant or RAG application. No user interaction, elevated privileges, or memory corruption primitives are required.

go
// Pre-patch vs patched interface in internal/app-service/client/client.go
// The fix threads userId and orgId through delete operations so that
// ownership is validated at the data access layer.

type IClient interface {
    // --- api key ---
    CreateApiKey(ctx context.Context, userId, orgId, name, desc string, expiredAt int64, apiKey string) (*model.OpenApiKey, *err_code.Status)
-   DeleteApiKey(ctx context.Context, keyId uint32) *err_code.Status
+   DeleteApiKey(ctx context.Context, keyId uint32, userId, orgId string) *err_code.Status
    UpdateApiKey(ctx context.Context, keyId uint32, userId, orgId, name, desc string, expiredAt int64) *err_code.Status

    // --- app key ---
    GetAppKeyList(ctx context.Context, userId, orgId, appId, appType string) ([]*model.ApiKey, *err_code.Status)
-   DelAppKey(ctx context.Context, appKeyId uint32) *err_code.Status
+   DelAppKey(ctx context.Context, appKeyId uint32, userId, orgId string) *err_code.Status
}

// internal/app-service/client/orm/api_key.go
-func (c *Client) DeleteApiKey(ctx context.Context, keyId uint32) *errs.Status {
-    if err := sqlopt.WithID(keyId).Apply(c.db.WithContext(ctx)).Delete(&model.OpenApiKey{}).Error; err != nil {
+func (c *Client) DeleteApiKey(ctx context.Context, keyId uint32, userId, orgId string) *errs.Status {
+    if err := sqlopt.SQLOptions(
+        sqlopt.WithID(keyId),
+        sqlopt.WithUserID(userId),
+        sqlopt.WithOrgID(orgId),
+    ).Apply(c.db.WithContext(ctx)).Delete(&model.OpenApiKey{}).Error; err != nil {
        return toErrStatus("api_key_delete", util.Int2Str(keyId), err.Error())
    }
    return nil
}

Source: GitHub commit 13d0b22

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 Attacks

Detection Methods for CVE-2026-108853

Indicators of Compromise

  • Repeated DELETE /v1/appspace/app requests from a single authenticated session targeting non-sequential or out-of-tenant appId values.
  • Sudden disappearance of tenant assistants, RAG applications, workflows, or conversation history without an authorized administrator action.
  • API audit entries where the deleting userId/orgId does not match the owning tenant of the deleted application.

Detection Strategies

  • Parse application access logs for DELETE operations under /v1/appspace/ and correlate the caller's tenant with the resource owner recorded in the database.
  • Build a rule that fires when a single principal issues more than a small threshold of delete requests across distinct appId values in a short window.
  • Diff backup snapshots against the live appspace tables to identify unexplained deletions that may indicate IDOR exploitation.

Monitoring Recommendations

  • Forward Wanwu bff-service and app-service logs to a centralized analytics pipeline with tenant context preserved.
  • Alert on HTTP 2xx responses to DELETE /v1/appspace/app when the request appId is not present in the caller's inventory listing.
  • Track EPSS movement for CVE-2026-108853 (currently 0.301%) alongside exposure of the Wanwu API to untrusted authenticated users.

How to Mitigate CVE-2026-108853

Immediate Actions Required

  • Upgrade all Wanwu deployments to version 0.6.3 or later, which introduces owner-scoped delete operations.
  • Restrict network access to the Wanwu API so that only trusted tenants can authenticate against /v1/appspace/app.
  • Audit recent delete activity for cross-tenant appId targeting and restore affected applications from backup where possible.

Patch Information

The fix is included in the v0.6.3 release and commit 13d0b225ec237d5a0b44730b2331efadaeb462a2. The patch threads userId and orgId through delete interfaces across app-service, api-key, and app-key flows, and enforces ownership in the SQL layer via sqlopt.WithUserID and sqlopt.WithOrgID. Additional analysis is available in the VulnCheck advisory and the HackMD CVE-2026-108853 analysis.

Workarounds

  • Place the Wanwu API behind a reverse proxy that enforces per-tenant allow-lists on DELETE /v1/appspace/app.
  • Temporarily disable self-service deletion for low-privileged roles and require administrator approval until the upgrade is applied.
  • Enable frequent database backups of the appspace tables so that destructive IDOR attempts can be reversed.
bash
# Pin Wanwu to the patched release and verify version
git clone https://github.com/UnicomAI/wanwu.git
cd wanwu
git checkout v0.6.3
grep -R "DeleteApiKey(ctx context.Context, keyId uint32, userId, orgId string)" internal/

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

Experience the Most Advanced Cybersecurity Platform

See how the world’s most intelligent, autonomous cybersecurity platform can protect your organization today and into the future.