CVE-2026-2532 is a server-side request forgery flaw in lintsinghua DeepAudit up to version 3.0.3. The embedding configuration endpoint handles IP addresses unsafely, and NVD scores it 6.3. The record says upgrading to 3.0.4 or 3.1.0 addresses it. That gives an attacker a way to make your server fetch addresses it should not.
What's affected
- Product: lintsinghua DeepAudit.
- Affected: versions up to and including 3.0.3.
- Fixed: 3.0.4 and 3.1.0, per the record.
- Component: the IP address handler in backend/app/api/v1/endpoints/embedding_config.py.
How to check if you're exposed
DeepAudit is a source code audit tool, so check the version you deployed. If you run it from a clone or a container, look at the tag or image you pulled:
git -C /path/to/DeepAudit describe --tags
docker image ls | grep -i deepaudit
Anything at 3.0.3 or below is in range. The project's issue 144 and pull request 145 describe the change.
What to do now
- Upgrade to 3.0.4 or 3.1.0.
- Restrict the service's outbound network access so it cannot reach cloud metadata addresses or internal admin hosts.
- Limit who can reach the embedding configuration API, and require authentication on it.
- Review logs for unexpected outbound requests from the DeepAudit host.
Why it matters
Server-side request forgery, CWE-918, means the server can be told to make a request to a place the caller chose. The danger is what the server can reach that the attacker cannot: internal services, metadata endpoints and admin panels.
The score is moderate because the vector expects a logged in user and limited impact. Still, SSRF in a tool that stores API settings for model providers deserves a quick patch. The record does not mark exploitation in the wild.
Quick check
How do I limit SSRF damage beyond patching?
Block egress to link local and private ranges from the host, and use network policy so the app can only reach what it needs.
Sources
Last reviewed: 19 October 2026.
