LMDeploy vulnerable to arbitrary code execution via eval() of untrusted quant_dtype in model config loading
Key facts
- Attack vector
- Network
- Privileges required
- None
- User interaction
- Required
- Attack complexity
- Low
- Impact type
- Denial of service
- Confidentiality · integrity · availability
- High · High · High
- Fix status
- Fixed
Attack profile from CVSS 3.1 · CNA. Fix status from CNA and GitHub. Impact type from CWE-400.
Summary
LMDeploy is a toolkit for compressing, deploying, and serving large language models. Versions 012.1 through 0.12.2 contain 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. Version 0.12.3 contains a patch.
Why it matters now
SSVC decision points
CISA ADP · SSVC 2.0.3
- Exploitation
- Poc
- Automatable
- No
- Technical impact
- Total
The decision (Track, Attend, Act) also depends on your mission and well-being impact; see CISA's SSVC guide.
Affected products
| Vendor | Product | Affected versions | Fixed in | Status | Source |
|---|---|---|---|---|---|
| InternLM | lmdeploy | 0.12.1 to before 0.12.3 | 0.12.3 | Patch available | CNA |
| — | lmdeploy (PyPI) | 0.12.1 to before 0.12.3 | 0.12.3 | Patch available | GHSA-3hmm-rh5q-gwwr |
- lmdeploy VendorInternLM Affected 0.12.1 to before 0.12.3 Fixed in 0.12.3 Patch available CNA
- lmdeploy (PyPI) Vendor— Affected 0.12.1 to before 0.12.3 Fixed in 0.12.3 Patch available GHSA-3hmm-rh5q-gwwr
Remediation
- Fix · CNA lmdeploy (InternLM) Fixed in 0.12.3
- Fix · GHSA-3hmm-rh5q-gwwr lmdeploy (PyPI) Fixed in 0.12.3 Advisory →
Timeline
- 2026-10-02 Threatist record created from CVE-2026-33625
- 2026-10-03 Alias added: GHSA-3hmm-rh5q-gwwr
References
More InternLM vulnerabilities
- TI-2026-67516 InternLM MindSearch Planner Agent graph.py ExecutionAction.run code injection CVSS 4.0CRITICAL 10.0
- TI-2025-37341 LMdeploy has Remote Code Execution by Pickle Deserialization via zmq_rpc.call_and_response() in InterLM/lmdeploy CVSS 3.1CRITICAL 9.8
- TI-2026-83181 Unauthenticated RCE in InternLM LMDeploy via pickle deserialization CVSS 4.0CRITICAL 9.3
- TI-2026-53684 LMDeploy Remote Code Execution via Unsafe Pickle Deserialization in the Disaggregated Serving Peer Connector CVSS 4.0CRITICAL 9.3
- TI-2026-62637 InternLM LMDeploy through 0.17.0 Memory Exhaustion via Session ID Mismatch CVSS 4.0HIGH 8.7
- TI-2026-62632 InternLM LMDeploy through 0.17.0 Assertion Denial of Service CVSS 4.0HIGH 8.7
Sources
- CVE-2026-33625 · CVE Program
- CVE-2026-33625 · NVD
- FIRST EPSS
- GHSA-3hmm-rh5q-gwwr · GitHub Advisory Database · CC BY 4.0
This product uses the NVD API but is not endorsed or certified by the NVD. Sources and licences