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-2026-108689 Overview
CVE-2026-108689 is a missing authorization vulnerability [CWE-639] in Wukong AICRM through build 20260610. The flaw resides in the POST /chat/send endpoint, which accepts a user-supplied sessionId without verifying ownership. Authenticated users can target arbitrary chat sessions belonging to other users. Attackers append messages to a victim's AI conversation and receive streamed assistant replies constructed from the victim's last 20 messages. The response discloses conversation content from the targeted session, breaching confidentiality of CRM interactions handled by the AI assistant.
Critical Impact
Any authenticated user can write to another user's AI chat session and read back streamed replies containing the victim's prior conversation context.
Affected Products
- Wukong AICRM through build 20260610
- ChatServiceImpl component in backend/src/main/java/com/kakarote/ai_crm/service/impl/
- Deployments exposing the /chat/send endpoint to authenticated users
Discovery Timeline
- 2026-10-11 - CVE-2026-108689 published to NVD
- 2026-10-13 - Last updated in NVD database
Technical Details for CVE-2026-108689
Vulnerability Analysis
Wukong AICRM exposes an AI chat API that stores conversation history per session. The POST /chat/send handler reads the sessionId directly from the client request body and uses it to append new messages and retrieve conversation context. The handler does not validate that the requesting user owns the referenced session. An authenticated attacker who enumerates or guesses a valid sessionId can submit a message that becomes part of another user's conversation history. The backend then forwards the last 20 messages of that session to the model and streams the generated reply back to the attacker. This leaks CRM conversation content, which may include customer data, sales context, or internal notes.
Root Cause
The root cause is a classic authorization bypass through user-controlled key [CWE-639]. The ChatServiceImpl logic trusts the sessionId parameter without cross-referencing it against the authenticated user's identity. Session ownership is treated as implicit rather than enforced server-side. Any valid authenticated token combined with a known or guessed session identifier satisfies the control.
Attack Vector
The vulnerability is reachable over the network by any authenticated user. The attacker sends a crafted POST /chat/send request specifying a victim's sessionId and arbitrary message content. The server appends the message, invokes the AI backend with the victim's accumulated history, and streams the assistant reply to the attacker. See the GitHub PoC Repository and the ChatService Implementation for technical details.
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-2026-108689
Indicators of Compromise
- Multiple distinct authenticated user accounts writing to the same sessionId in application logs.
- POST /chat/send requests where the authenticated user does not match the owner recorded in the session table.
- Unexpected spikes in streamed chat completions tied to sessions that have not been opened in the UI by their owner.
Detection Strategies
- Instrument the /chat/send endpoint to log userId, sessionId, and session owner, then alert on ownership mismatches.
- Correlate chat session access patterns against UI session-open events to identify API-only writers.
- Review database audit trails for message inserts attributed to users who never created the parent session.
Monitoring Recommendations
- Monitor outbound streamed responses for abnormal volume per authenticated user.
- Track enumeration-style request patterns against /chat/send with sequential or randomized sessionId values.
- Alert on authenticated accounts generating conversation reads significantly exceeding their historical baseline.
How to Mitigate CVE-2026-108689
Immediate Actions Required
- Restrict access to the /chat/send endpoint behind an application-layer check that validates sessionId ownership against the authenticated principal.
- Rotate or expire active chat sessions to invalidate any identifiers that may have been enumerated.
- Audit historical logs for cross-user sessionId usage and notify affected tenants of potential disclosure.
Patch Information
At publication time, the VulnCheck Advisory on Wukong AICRM does not list a fixed version. Operators should track the upstream WuKongOpenSource repository for a patched build superseding 20260610 and deploy it once published.
Workarounds
- Add a server-side authorization filter that rejects /chat/send requests when the session owner differs from the authenticated user.
- Replace client-supplied sessionId values with a server-resolved mapping keyed to the authenticated identity.
- Place the AI chat API behind a reverse proxy policy that enforces per-user session scoping until an official fix ships.
# Example reverse-proxy guard: reject chat/send if sessionId cookie
# does not match server-issued ownership token
location /chat/send {
if ($http_x_session_owner != $cookie_user_id) {
return 403;
}
proxy_pass http://aicrm_backend;
}
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

