CVE-2026-130358: AdminNeo AI plugins expose operator API keys to unauthenticated attackers

AdminNeo before 5.9.1 contains a missing authentication vulnerability in three AI SQL-helper plugins that allows unauthenticated attackers to relay arbitrary prompts using the operator's API credentials.

Christopher Ort
By Christopher Ort · 3 min read
Share
A newspaper photograph showing a database administrator looking concerned at a computer screen displaying authentication logs with unauthorized access attempts
Image: AI-generated, i10X
On this page
  1. Quick Take
  2. In plain English
  3. What happened
  4. What you should do
  5. What it means
  6. Sources
i10X Quick Take

Quick Take

  1. What: CVE-2026-130358, a missing authentication vulnerability in AdminNeo’s ClaudeSqlPlugin, GeminiSqlPlugin, and OpenWebUiPlugin that allows unauthenticated attackers to relay arbitrary prompts using the operator’s API keys or bearer tokens
  2. Who is affected: AdminNeo installations before version 5.9.1 that have enabled any of the three AI SQL-helper plugins (ClaudeSqlPlugin, GeminiSqlPlugin, or OpenWebUiPlugin)
  3. Severity: Medium (CVSS 6.5), allows unauthenticated access to paid AI accounts and potential exposure of internal endpoints
  4. Status: Fixed in AdminNeo 5.9.1 released November 16, 2026
  5. What to do: Upgrade to AdminNeo 5.9.1 or later if you have enabled any of the affected AI plugins

In plain English

AdminNeo is a database management tool that can connect to AI services like Claude, Gemini, or Open WebUI to help generate SQL queries. Three plugins that provide this AI assistance had a serious flaw: they would accept and forward prompts to these AI services even from visitors who had not logged in. This means anyone who could reach the AdminNeo login page could use the administrator’s paid AI account to send any prompts they wanted. The developers fixed this in version 5.9.1 by requiring authentication before the plugins will work.

What you should do: If you use AdminNeo with any of these AI plugins enabled, you should upgrade immediately, otherwise you can ignore this issue.

What happened

AdminNeo versions before 5.9.1 contained an authentication flaw in three plugins designed to help generate SQL queries using AI services: ClaudeSqlPlugin, GeminiSqlPlugin, and OpenWebUiPlugin. These plugins performed their work in a sendHeaders() function that executed on every page load, including the login page shown to unauthenticated visitors.

When an attacker sent a POST request with a claude, gemini, or openwebui field to the AdminNeo login page, the corresponding plugin would forward that text to the configured AI endpoint using the operator’s credentials (x-api-key, API key, or Authorization bearer token) and return the model’s response. The plugins never checked whether the user was authenticated and never verified the CSRF token.

A security researcher verified the vulnerability end-to-end by configuring OpenWebUiPlugin to point at an instrumented local endpoint with a dummy bearer token. Sending a POST request with no session cookie or CSRF token successfully relayed the prompt to the endpoint with the operator’s bearer token in the Authorization header, and the response was returned to the unauthenticated caller.

The vulnerability affected all releases shipping these plugins. OpenWebUiPlugin and GeminiSqlPlugin have been vulnerable since their introduction, while ClaudeSqlPlugin has been vulnerable since commit 4111cac1. Each plugin must be explicitly enabled by the operator for the vulnerability to be exploitable.

What you should do

AdminNeo users who have enabled ClaudeSqlPlugin, GeminiSqlPlugin, or OpenWebUiPlugin should upgrade to version 5.9.1 immediately. The fixed version is available from the project download page and GitHub releases.

Administrators should review their AI service billing and access logs for any unexpected usage that may indicate exploitation. For OpenWebUiPlugin users, check logs on internal Open WebUI endpoints for unauthorized access attempts.

If immediate upgrade is not possible, disable the affected AI plugins until the upgrade can be completed.

What it means

This vulnerability is a bypass of authentication controls for a sensitive feature. Any unauthenticated visitor who could reach the AdminNeo URL could use it as an open proxy to the operator’s AI account, sending arbitrary prompts that would be billed to the operator with full responses returned.

For OpenWebUiPlugin, the impact extends beyond unauthorized API usage. Because Open WebUI endpoints are frequently hosted on internal networks, the vulnerability also provided unauthenticated access to internal services using the operator’s bearer token, crossing the trust boundary between unauthenticated external actors and privileged internal credentials.

The CVSS score of 6.5 (Medium severity) reflects the network-accessible attack vector requiring no privileges or user interaction, with low impact to integrity and availability. However, organizations using these plugins in production should treat this as a high-priority patch due to the potential for credential exposure and unauthorized access to internal systems.

The vulnerability was assigned CVE-2026-130358 and is tracked as GHSA-cjvx-f3xp-vg55. It falls under CWE-306 (Missing Authentication for Critical Function) and CWE-862 (Missing Authorization).

Exposed AI provider keys are a recurring pattern: see how DeepAudit exposed LLM API keys and Git tokens and the pgEdge RAG Server leaked LLM provider API keys in error messages. New cases are tracked in our AI Security hub.

Sources

GitHub Security Advisory GHSA-cjvx-f3xp-vg55 (adminneo-org/adminneo); AdminNeo 5.9.1 release notes; NVD entry CVE-2026-130358.

Christopher Ort

Christopher Ort

Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU). He is the author of “The Style Penalty”, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.

Editor's picks

  1. Genesys Launches Standalone AI Agents and Arena, With Pricing Based on How Hard Each Task Is
  2. Seedance 2.5 4K Reaches Developers: 10-Bit Video up to 30 Seconds With 480p Draft Mode
  3. Perplexity Computer Adds Multiagent Mode, Where Up to 16 Agents on Rival Models Check Each Other
  4. Gemini Can Now Make Calls, Pay Bills and Shop, Free in Six Countries With Gemini 4 Hydrogen
  5. Hinton on Firing Line: AI Is Making ‘Alien Beings’ and Self-Regulation Is ‘Absurd’