Safeguard
Vulnerability Analysis•

CVE-2025-11046: High SSRF Flaw in Tencent WeKnora

CVE-2025-11046 is a high severity SSRF in Tencent WeKnora 0.1.0 via the embedding test endpoint. No fix is listed: check releases and restrict API access.

Safeguard Security Team
3 min read

CVE-2025-11046 is a high severity (CVSS 7.3) server-side request forgery flaw in Tencent WeKnora 0.1.0. The baseUrl argument of the embedding test endpoint can be abused remotely. The record names no fixed release, so check the project's releases, and keep the endpoint away from untrusted users.

What's affected

The record names Tencent WeKnora 0.1.0. The function is testEmbeddingModel, served at /api/v1/initialization/embedding/test, and the baseUrl argument is the one that matters. NVD scores it 7.3 (HIGH) with the vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:L. Weakness class: CWE-918.

No fixed version is named in the record. Check the WeKnora releases and the repository for a later version before you decide.

How to check if you're exposed

Find out how WeKnora is deployed. If you run containers, list the images and tags. If you built from source, read the version from the checkout.

docker ps --format '{{.Image}}' | grep -i weknora
git -C /path/to/WeKnora describe --tags

Then check whether /api/v1/initialization/embedding/test is reachable by people who should not have access. Review your reverse proxy and firewall rules for it.

What to do now

  1. Find every WeKnora deployment and its version.
  2. Check the project's releases for a build later than 0.1.0 that addresses the issue, and read its notes before you upgrade.
  3. Put the API behind authentication and a firewall so only trusted users reach the initialization endpoints.
  4. Limit outbound traffic from the server so it cannot reach internal services or cloud metadata addresses.
  5. Review logs for calls to the embedding test endpoint with unexpected base URLs.

Why it matters

Server-side request forgery lets a user make the server send requests on their behalf. The server often sits inside the network and can reach places the user cannot, such as internal admin pages or cloud metadata services. Here the base URL of an embedding model test is the lever.

Tools that connect to model providers need to accept URLs, so the fix is not to ban them. It is to validate where they point, restrict egress, and keep setup endpoints away from anonymous users. The weakness is CWE-918.

Try it yourself

Click through it at your own pace, or open it full screen.

Quick check

What is the endpoint to protect?

The record names the embedding test endpoint under /api/v1/initialization/embedding/test and its baseUrl argument.

Does limiting outbound traffic help?

Yes. If the server cannot reach internal services or metadata addresses, a forged request has far less to find.

Sources

Last reviewed: 19 October 2026.

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

We email you a link to confirm, and only then add you to the list. Unsubscribe in one click. Privacy

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.