CVE-2026-12203: HKUDS AI-Trader Info Disclosure Flaw

CVE-2026-12203 Overview

CVE-2026-12203 is an information disclosure vulnerability in HKUDS AI-Trader, an open-source AI-driven trading platform. The flaw exists in the /api/research/agents.csv endpoint of the Research Export component. An unauthenticated remote attacker can request the endpoint and retrieve sensitive research data without supplying credentials. The vendor confirmed that research export endpoints previously lacked authentication enforcement and now require an authenticated agent with the research_exports capability. The exploit details have been published, increasing the risk of opportunistic abuse against exposed instances. This issue is tracked under CWE-200: Exposure of Sensitive Information to an Unauthorized Actor.

Critical Impact
Unauthenticated remote attackers can extract sensitive research agent data from exposed AI-Trader deployments via a single HTTP request to the /api/research/agents.csv endpoint.

Affected Products

HKUDS AI-Trader up to commit 74caf996f78dcc0c657df8365c8544678a16e215
Research Export component (/api/research/agents.csv endpoint)
Rolling release builds prior to patch commit 91a31aac1b0f4dbc6b8bef9f6eff0b7912e0bc65

Discovery Timeline

2026-06-15 - CVE-2026-12203 published to NVD
2026-06-17 - Last updated in NVD database

Technical Details for CVE-2026-12203

Vulnerability Analysis

The vulnerability resides in the Research Export functionality of HKUDS AI-Trader. The endpoint /api/research/agents.csv returns research agent data formatted as CSV without enforcing authentication or authorization checks. Any remote client capable of reaching the service over the network can retrieve the exported data.

The attack requires no privileges, no user interaction, and low complexity. Because the project follows a rolling release model, no semantic version range applies. Instead, the vulnerable state is defined by the commit history up to 74caf996f78dcc0c657df8365c8544678a16e215.

The public availability of the exploit details, combined with the simplicity of an HTTP GET request, makes this an attractive target for automated scanning against internet-exposed AI-Trader instances.

Root Cause

The root cause is missing access control on a sensitive data export route. The Research Export endpoint did not validate the requesting client or check for a capability token before returning research agent data. The vendor remediation introduces an authentication requirement gated by a new research_exports capability, aligning the endpoint with standard authorization patterns applied elsewhere in the API.

Attack Vector

An attacker sends an HTTP GET request to the Research Export endpoint on a reachable AI-Trader deployment. The server returns the CSV payload containing research agent data without verifying any session or bearer token. The patch demonstrates the corrective pattern, where API calls now include an Authorization: Bearer header and the server rejects unauthenticated requests with an error.

text

   const loadExperiments = async () => {
     setLoading(true)
     try {
-      const res = await fetch(`${API_BASE}/experiments?limit=100`)
+      const res = await fetch(`${API_BASE}/experiments?limit=100`, {
+        headers: token ? { 'Authorization': `Bearer ${token}` } : {}
+      })
       const data = await res.json()
+      if (!res.ok) throw new Error(data.detail || 'experiment_load_failed')
       setExperiments(data.experiments || [])

Source: GitHub Commit 91a31aa — patch fragment from service/frontend/src/ExperimentAdminPage.tsx showing the introduction of bearer token authorization and error handling on protected API calls.

Detection Methods for CVE-2026-12203

Indicators of Compromise

Unauthenticated HTTP GET requests to /api/research/agents.csv originating from external IP addresses.
HTTP 200 responses with Content-Type: text/csv returned to clients without prior authentication exchanges.
Repeated automated scanning patterns hitting research export paths across multiple AI-Trader instances.

Detection Strategies

Inspect reverse proxy and application access logs for requests to /api/research/agents.csv that lack an Authorization header.
Correlate CSV export downloads with active session activity to identify anonymous exports.
Compare the deployed AI-Trader commit hash against the patched commit 91a31aac1b0f4dbc6b8bef9f6eff0b7912e0bc65 to confirm exposure.

Monitoring Recommendations

Enable verbose API access logging on the AI-Trader service and forward logs to a centralized analytics pipeline.
Alert on any 200 response from research export endpoints when no authenticated agent context is present in the request.
Monitor outbound bandwidth on the AI-Trader host for unusual CSV download volumes that may indicate bulk extraction.

How to Mitigate CVE-2026-12203

Immediate Actions Required

Pull and deploy the patched code at commit 91a31aac1b0f4dbc6b8bef9f6eff0b7912e0bc65 or later from the HKUDS AI-Trader repository.
Restrict network exposure of the AI-Trader API to trusted networks or an authenticated reverse proxy until the patch is applied.
Audit historical access logs for prior unauthenticated requests to /api/research/agents.csv and treat returned data as potentially exposed.

Patch Information

The upstream fix is delivered as commit 91a31aac1b0f4dbc6b8bef9f6eff0b7912e0bc65, merged via GitHub Pull Request #227 and tracking GitHub Issue #242. The patch enforces that Research Export endpoints require an authenticated agent holding the research_exports capability. Because AI-Trader uses rolling releases, operators must redeploy from source rather than upgrade a versioned package. Additional context is available in the GitHub Advisory and on VulDB CVE-2026-12203.

Workarounds

Place AI-Trader behind a reverse proxy that enforces authentication on /api/research/* routes until patching is complete.
Block external access to research export paths at the web application firewall or load balancer layer.
Disable or remove the research export feature in deployments where it is not required.

bash

# Nginx example: require authentication on research export endpoints
location ~ ^/api/research/.*\.csv$ {
    auth_request /auth-validate;
    proxy_pass http://ai_trader_backend;
}

location = /auth-validate {
    internal;
    proxy_pass http://auth_service/validate;
    proxy_pass_request_body off;
    proxy_set_header Content-Length "";
    proxy_set_header X-Original-URI $request_uri;
}