CVE-2026-58116
LLaMA-Factory through 0.9.5 contains a remote code execution vulnerability that allows attackers with WebUI access to execute arbitrary Python code by supplying a malicious model path in the Chat or Training interfaces.
Does this matter?
High impact if exploited, but EPSS currently rates exploitation as unlikely (0.90%). Schedule it in the normal patch cycle and watch for a rise in EPSS or a public exploit.
Description
LLaMA-Factory through 0.9.5 contains a remote code execution vulnerability that allows attackers with WebUI access to execute arbitrary Python code by supplying a malicious model path in the Chat or Training interfaces. The application passes user-supplied model path input unvalidated into AutoTokenizer.from_pretrained() and AutoModel.from_pretrained() with a hardcoded trust_remote_code=True parameter, causing the Hugging Face transformers library to fetch and execute arbitrary code from a remote or local model repository with the privileges of the server process.
- CVSS 4.0
- 9.3 CRITICALCVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/M
- EPSS
- 0.90% probability · 58th percentile
- CISA KEV
- Not listed
- Weakness
- CWE-94, CWE-829
- Affected
- hiyouga/llama-factory
- Source
- disclosure@vulncheck.com
References
- https://gist.github.com/henrrrychau/08d76ec672f42136bbc1449c4f2973f8Exploit, Mitigation, Third Party Advisory
- https://www.vulncheck.com/advisories/llama-factory-remote-code-execution-via-webui-model-pathThird Party Advisory
Source: NVD record, EPSS from FIRST.org, KEV from CISA. Refreshed daily.