VAITP Dataset

← Back to the dataset

CVE-2026-58474

Code injection via crafted GGUF filename in whichllm 0.5.16 CLI commands.

  • CVSS 8.6
  • 94
  • Input Validation and Sanitization
  • Remote

whichllm before 0.5.16 contains a code injection vulnerability in the run and snippet commands that allows a remote attacker who controls a HuggingFace repository to achieve arbitrary code execution by crafting a malicious GGUF filename containing double quotes or other special characters. The script generation function in cli.py interpolates HuggingFace-derived values, including GGUF variant filenames from the Hub API siblings rfilename field, directly into Python source code without escaping, allowing the crafted filename to break out of the generated string literal and execute injected code on the user's machine before any model download occurs.

CWE
94
CVSS base score
8.6
Published
2026-08-26
OWASP
A03 Injection
Orthogonal defect classification
Checking
Code defect classification
Missing Check
Category
Input Validation and Sanitization
Subcategory
Command Injection
Accessibility scope
Remote
Impact
Arbitrary Code Execution
Affected component
whichllm
Fixed by upgrading
Yes

Solution

Upgrade to whichllm ≥ 0.5.16.

Vulnerable code sample

import sys
import json
from huggingface_hub import HfApi

def generate_script(repo_id: str, variant: str) -> str:
    api = HfApi()
    # Retrieve sibling filenames for the given variant
    siblings = api.list_repo_files(repo_id, repo_type="model")
    # Find the GGUF file matching the variant
    gguf_file = next((s for s in siblings if s.endswith(f"-{variant}.gguf")), None)
    if not gguf_file:
        raise FileNotFoundError("Variant not found")
    # VULNERABLE: directly interpolating untrusted filename into generated code
    script = f'''
import torch
model = torch.load("{gguf_file}", map_location="cpu")
print("Model loaded")
'''
    return script

Patched code sample

import sys
import json
import ast
from huggingface_hub import HfApi

def generate_script(repo_id: str, variant: str) -> str:
    api = HfApi()
    siblings = api.list_repo_files(repo_id, repo_type="model")
    gguf_file = next((s for s in siblings if s.endswith(f"-{variant}.gguf")), None)
    if not gguf_file:
        raise FileNotFoundError("Variant not found")
    # FIX: safely embed the filename using ast.literal_eval to ensure proper escaping
    safe_name = ast.literal_eval(repr(gguf_file))
    script = f'''
import torch
model = torch.load({safe_name}, map_location="cpu")
print("Model loaded")
'''
    return script

Payload

__VAITP_MODEL_REFUSED__

Cite this entry

@misc{vaitp:cve202658474,
  title        = {{Code injection via crafted GGUF filename in whichllm 0.5.16 CLI commands.}},
  author       = {Bogaerts, Fr\'ed\'eric and Ivaki, Naghmeh and Fonseca, Jos\'e},
  year         = {2026},
  note         = {VAITP Python Vulnerability Dataset, entry CVE-2026-58474},
  howpublished = {\url{https://netpack.pt/vaitp/vulnerability/CVE-2026-58474/}}
}
Introducing the "VAITP dataset": a specialized repository of Python vulnerabilities and patches, meticulously compiled for the use of the security research community. As Python's prominence grows, understanding and addressing potential security vulnerabilities become crucial. Crafted by and for the cybersecurity community, this dataset offers a valuable resource for researchers, analysts, and developers to analyze and mitigate the security risks associated with Python. Through the comprehensive exploration of vulnerabilities and corresponding patches, the VAITP dataset fosters a safer and more resilient Python ecosystem, encouraging collaborative advancements in programming security.

The supreme art of war is to subdue the enemy without fighting.

Sun Tzu – “The Art of War”

:: Shaping the future through research and ingenuity ::