CVE-2026-123818: LMDeploy URL validation bypass via backslash interpretation

CVE-2026-123818 · Severity: high · CVSS 7.5 · Published 2026-11-09

LMDeploy is a toolkit for deploying and serving large language models that accepts remote media URLs from users. An attacker can craft a URL that passes validation but resolves to an internal server when actually fetched, allowing access to internal HTTP services and sensitive data. This affects versions 0.12.3 through 0.15.0.

Technologies: Internlm Lmdeploy. Vendors: Internlm.

Quick answers

Is CVE-2026-123818 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2026-123818 being exploited in the wild.
How severe is CVE-2026-123818?
High severity, with a CVSS score of 7.5, published 9 November 2026.
Which products and versions are affected?
Internlm Lmdeploy 0.12.3 to 0.15.0.
Is there a fix for CVE-2026-123818?
Yes: fixed in 0.15.0. Until then: Use the --allowed-media-domains flag at API server startup to restrict which hostnames HTTP(S) media URLs can fetch from. This enforces exact hostname matching and significantly reduces SSRF risk for deployments exposed to untrusted clients.</anigation> <parameter name="timeline">[{"kind":"disclosed","at":"2026-11-09"},{"kind":"patched","at":"2026-07-20"}] Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.
How is CVE-2026-123818 attacked?
It is attacked over the network, in a software library.
Am I affected by CVE-2026-123818?
Check whether you run Internlm Lmdeploy at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

How to fix

  • Fixed in: 0.15.0

Use the --allowed-media-domains flag at API server startup to restrict which hostnames HTTP(S) media URLs can fetch from. This enforces exact hostname matching and significantly reduces SSRF risk for deployments exposed to untrusted clients.</anigation> <parameter name="timeline">[{"kind":"disclosed","at":"2026-11-09"},{"kind":"patched","at":"2026-07-20"}]

Technical details

The _is_safe_url function validates remote media URLs using Python's urlparse, but the requests library interprets backslashes differently when sending HTTP requests, allowing hostname validation to be bypassed. An attacker can provide a URL that appears to target a globally routable host during validation but whose backslash-based encoding resolves to an internal/private IP address during the actual fetch, bypassing non-global address checks. This is a Server-Side Request Forgery (SSRF) vulnerability affecting applications that accept untrusted remote media URLs.

Affected products

  • InternLM LMDeploy 0.12.3 to 0.15.0

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.