{"id":2377,"cve":"CVE-2026-33625","cve_url":"https:\/\/www.cvedetails.com\/cve\/CVE-2026-33625\/","cwe":"CWE-400","cwe_number":400,"cwe_url":"https:\/\/cwe.mitre.org\/data\/definitions\/400","cvss_score":8.8,"published":"2026-09-18","owasp":"A03 Injection","odc":"Checking","cdc":"Missing Check","category":"Input Validation and Sanitization","subcategory":"Command Injection","scope":"Remote","impact":"Arbitrary Code Execution","summary":"Arbitrary code exec via eval of unvalidated quant_dtype in lmdeploy config","permalink":"https:\/\/netpack.pt\/vaitp\/vulnerability\/CVE-2026-33625\/","description":"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.","solution":"Upgrade LMDeploy to version\u202f0.12.3\u202for later.","affected_component":"lmdeploy","fixed_by_upgrading":true,"vulnerable_code":"import torch\n\ndef get_quant_dtype(config):\n    # Extract quant_dtype from config\n    quant_cfg = config.get('quantization_config', {})\n    quant_dtype = quant_cfg.get('quant_dtype', 'float32')\n    # VULNERABLE: unsafe eval of user-controlled string\n    dtype = eval(f'torch.{quant_dtype}')\n    return dtype\n\ndef load_model(model_path, config):\n    dtype = get_quant_dtype(config)\n    # Placeholder for actual model loading logic\n    model = torch.nn.Module()\n    model.dtype = dtype\n    return model\n","patched_code":"import torch\n\nALLOWED_DTYPES = {'float32', 'float16', 'bfloat16'}\n\ndef get_quant_dtype(config):\n    # Extract quant_dtype from config\n    quant_cfg = config.get('quantization_config', {})\n    quant_dtype = quant_cfg.get('quant_dtype', 'float32')\n    # FIX: whitelist validation instead of eval\n    if quant_dtype not in ALLOWED_DTYPES:\n        raise ValueError(f'Unsupported quant_dtype: {quant_dtype}')\n    dtype = getattr(torch, quant_dtype)\n    return dtype\n\ndef load_model(model_path, config):\n    dtype = get_quant_dtype(config)\n    # Placeholder for actual model loading logic\n    model = torch.nn.Module()\n    model.dtype = dtype\n    return model\n","payload_needed":true,"payload":null,"payload_status":"model_refused","published_fix":[],"references":null}