Introduction
Understanding Static Application Security Testing
SAST helps in:
1) Finding the weaknesses in the code that might materialize into vulnerabilities. 2) Providing recommendations for fixing the weaknesses.
Downloading the Source Code
1
git clone https://gitlab.practical-devsecops.training/marudhamaran/caisp-image-classifier.git && cd caisp-image-classifier
Installing Bandit
1
pip install bandit==1.8.3
Usage:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
bandit -r .
[main] INFO profile include tests: None
[main] INFO profile exclude tests: None
[main] INFO cli include tests: None
[main] INFO cli exclude tests: None
[main] INFO running on Python 3.10.12
Run started:2025-11-28 05:01:02.086392
Test results:
>> Issue: [B614:pytorch_load] Use of unsafe PyTorch load
Severity: Medium Confidence: High
CWE: CWE-502 (https://cwe.mitre.org/data/definitions/502.html)
More Info: https://bandit.readthedocs.io/en/1.8.3/plugins/b614_pytorch_load.html
Location: ./image-classifier.py:113:30
112 # load best model weights
113 model.load_state_dict(torch.load(best_model_params_path))
114 return model
--------------------------------------------------
>> Issue: [B614:pytorch_load] Use of unsafe PyTorch load
Severity: Medium Confidence: High
CWE: CWE-502 (https://cwe.mitre.org/data/definitions/502.html)
More Info: https://bandit.readthedocs.io/en/1.8.3/plugins/b614_pytorch_load.html
Location: ./image-classifier.py:160:20
159 if torch.__version__[:3] == '2.3':
160 model = torch.load(model_file)
161 else:
--------------------------------------------------
>> Issue: [B614:pytorch_load] Use of unsafe PyTorch load
Severity: Medium Confidence: High
CWE: CWE-502 (https://cwe.mitre.org/data/definitions/502.html)
More Info: https://bandit.readthedocs.io/en/1.8.3/plugins/b614_pytorch_load.html
Location: ./image-classifier.py:175:24
174 ):
175 model = torch.load(model_file)
176 else:
--------------------------------------------------
Code scanned:
Total lines of code: 140
Total lines skipped (#nosec): 0
Run metrics:
Total issues (by severity):
Undefined: 0
Low: 0
Medium: 3
High: 0
Total issues (by confidence):
Undefined: 0
Low: 0
Medium: 0
High: 3
Files skipped (0):
Bandit’s results indicate that:
There are 3 weaknesses in the source code. 3 of the weaknesses are medium severity weaknesses. Bandit is pretty confident about the results, indicating that those three weaknesses are reported with high confidence.
The three issues were reported in the following line numbers in the image-classifier.py file:
113 160 175 The code present around those lines were also shown, but we did not include them in the above output for brevity.
The three issues are related to the use of torch.load function.
The bandit tool has provided the following links to help us understand the issue, and how to fix it:
https://bandit.readthedocs.io/en/1.8.3/plugins/b614_pytorch_load.html https://cwe.mitre.org/data/definitions/502.html Let’s navigate to the first link, and understand the issue - https://bandit.readthedocs.io/en/1.8.3/plugins/b614_pytorch_load.html.
To fix, the documentation recommends the following:
1
2
3
There are two safe alternatives:
1. Use torch.load with weights_only=True where only tensor data is extracted, and no arbitrary Python objects are deserialized
2. Use the safetensors library from huggingface, which provides a safe deserialization mechanism
With weights_only=True, PyTorch enforces a strict type check, ensuring that only torch.Tensor objects are loaded. So, it appears that the torch.load function is not safe to use, the way it is currently used in the code.
And, it also appears that the torch.load function also takes another parameter called weights_only, which when set to True, ensures that only tensor data is loaded, and no arbitrary Python objects are deserialized.
Let’s go ahead and fix the issue by adding the weights_only parameter to the torch.load function.
You can open the image-classifier.py file in vi or nano editor, and add the weights_only parameter to the torch.load function, set to True.
Or you can use the below sed command to that in an easy way.
1
2
sed -i "s/torch.load(\(best_model_params_path\))/torch.load(\1, weights_only=True)/g" image-classifier.py
sed -i "s/torch.load(\(model_file\))/torch.load(\1, weights_only=True)/g" image-classifier.py