Description
LLaMA-Factory contains a server-side request forgery vulnerability in the OpenAI-compatible API multimodal media URL handler that allows unauthenticated attackers to bypass SSRF validation. The check_ssrf_url guard validates URLs once but requests.get follows redirects and re-resolves DNS without re-validation, enabling attackers to use HTTP redirects or DNS rebinding to access internal addresses and cloud metadata endpoints.
Published: 2026-09-04
Score: 8.7 High
EPSS: < 1% Very Low
KEV: No
Impact: Server‑Side Request Forgery allowing internal network access
Action: Immediate Patch
AI Analysis

Impact

LLaMA‑Factory contains a server‑side request forgery in the OpenAI‑compatible multimodal media URL handler where the guard validates a URL only once, but subsequent HTTP requests follow redirects and resolve DNS again without re‑validation; an unauthenticated attacker can supply a malicious URL that redirects or uses DNS rebinding to target internal or cloud metadata endpoints, exposing sensitive information or enabling further attacks, a flaw classified as CWE-918 that directly threatens confidentiality and availability.

Affected Systems

The vulnerability applies to the hiyouga:LlamaFactory product, affecting any installation that has not moved beyond version 0.9.5; the advisory references a 0.9.5 repository and no additional version qualifiers are provided.

Risk and Exploitability

With a CVSS score of 8.7 and no EPSS data available, the flaw is high severity though exploitation likelihood is uncertain; it is unauthenticated, so any HTTP request to the affected endpoint can trigger it, potentially allowing attackers to access internal resources, leak data, or proceed with further compromise; the CVE is not listed in the CISA KEV catalog but remains a priority for exposed deployments.

Generated by OpenCVE AI on September 4, 2026 at 15:49 UTC.

Remediation

No vendor fix or workaround currently provided.

OpenCVE Recommended Actions

  • Apply an official patch or upgrade to the latest LLaMA‑Factory release that revalidates URLs after redirects and DNS resolution.
  • Configure network controls or firewall rules to block outbound connections to internal IP ranges and metadata service endpoints from the LLaMA‑Factory container.
  • Disable or restrict the multimodal media URL processing feature in the OpenAI‑compatible API until a secure fix is applied.

Generated by OpenCVE AI on September 4, 2026 at 15:49 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories yet.

History

Fri, 04 Sep 2026 16:30:00 +0000

Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'yes', 'Exploitation': 'poc', 'Technical Impact': 'partial'}, 'version': '2.0.3'}


Fri, 04 Sep 2026 14:45:00 +0000

Type Values Removed Values Added
Description LLaMA-Factory contains a server-side request forgery vulnerability in the OpenAI-compatible API multimodal media URL handler that allows unauthenticated attackers to bypass SSRF validation. The check_ssrf_url guard validates URLs once but requests.get follows redirects and re-resolves DNS without re-validation, enabling attackers to use HTTP redirects or DNS rebinding to access internal addresses and cloud metadata endpoints.
Title LLaMA-Factory SSRF Guard Bypass via Redirect and DNS Rebinding
First Time appeared Hiyouga
Hiyouga llama-factory
Weaknesses CWE-918
CPEs cpe:2.3:a:hiyouga:llama-factory:*:*:*:*:*:*:*:*
Vendors & Products Hiyouga
Hiyouga llama-factory
References
Metrics cvssV3_1

{'score': 7.5, 'vector': 'CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N'}

cvssV4_0

{'score': 8.7, 'vector': 'CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:N/VA:N/SC:N/SI:N/SA:N'}


Subscriptions

Hiyouga Llama-factory
cve-icon MITRE

Status: PUBLISHED

Assigner: VulnCheck

Published:

Updated: 2026-09-04T15:19:48.538Z

Reserved: 2026-09-04T13:32:37.698Z

Link: CVE-2026-85673

cve-icon Vulnrichment

Updated: 2026-09-04T15:18:42.894Z

cve-icon NVD

Status : Received

Published: 2026-09-04T15:17:45.440

Modified: 2026-09-04T16:18:22.060

Link: CVE-2026-85673

cve-icon Redhat

No data.

cve-icon OpenCVE Enrichment

Updated: 2026-09-04T16:00:05Z

Weaknesses
  • CWE-918

    Server-Side Request Forgery (SSRF)