CVE-2026-33626 LMDeploy Vulnerable to Server-Side Request Forgery (SSRF) via Vision-Language Image Loading
Attackers can exploit SSRF in LMDeploy's vision module to access internal networks and cloud metadata by providing malicious URLs.
- Exploited in the wild
- Not known
- Login needed
- Not needed
- Attackable
- Over the internet
- Victim action needed
- Not needed
- Fix
- None listed
What is CVE-2026-33626?
CVE-2026-33626 is a high-severity server-side request forgery vulnerability in InternLM lmdeploy, published on 20 Apr 2026. An attacker can exploit it over the internet without logging in, with no action from the victim.
Official description
LMDeploy is a toolkit for compressing, deploying, and serving large language models. Versions prior to 0.12.3 have a Server-Side Request Forgery (SSRF) vulnerability in LMDeploy's vision-language module. The `load_image()` function in `lmdeploy/vl/utils.py` fetches arbitrary URLs without validating internal/private IP addresses, allowing attackers to access cloud metadata services, internal networks, and sensitive resources. Version 0.12.3 patches the issue.
Weakness: CWE-918
Which versions are affected by CVE-2026-33626?
3 open source packages ship an affected version, including lmdeploy. It affects 1 product.
| Package | Affected | Fixed in |
|---|---|---|
lmdeploy PyPI | All versions to 0.12.2 | None |
lmdeploy PyPI | All versions to 0.12.2 | None |
lmdeploy PyPI | All versions to 0.12.2 | None |
Products include Internlm Lmdeploy.
How do I fix CVE-2026-33626?
No fix has been published for CVE-2026-33626 yet. Limit exposure of the affected software until one is available.
Upgrade, and send outbound requests through an allow-list that blocks internal addresses. More about Server-side request forgery.
Is CVE-2026-33626 being exploited?
It is not confirmed as exploited in the wild, but public exploit code for CVE-2026-33626 exists.
Public exploit code: 1 Nuclei template and 1 proofs of concept on GitHub.
Other vulnerabilities in InternLM lmdeploy
- CVE-2025-67729 lmdeploy vulnerable to Arbitrary Code Execution via Insecure Deserialization in torch.load()
- CVE-2026-63764 LMDeploy Server-Side Request Forgery via HTTP Redirect Bypass
- CVE-2025-3162 InternLM LMDeploy PT File utils.py load_weight_ckpt deserialization
- CVE-2025-3163 InternLM LMDeploy conf.py open code injection
Sources
Also known as CVE-2025-62167, GHSA-6w67-hwm5-92mq, PYSEC-2026-2607.