Skip to content

Instantly share code, notes, and snippets.

@shaoyuyoung
Last active September 24, 2025 02:41
Show Gist options
  • Select an option

  • Save shaoyuyoung/0e7d2a586297ae9c8ed14d8706749efc to your computer and use it in GitHub Desktop.

Select an option

Save shaoyuyoung/0e7d2a586297ae9c8ed14d8706749efc to your computer and use it in GitHub Desktop.
CVE-2025-55551, CVE-2025-55552, CVE-2025-55553, CVE-2025-55554, CVE-2025-55556, CVE-2025-55557, CVE-2025-55558, CVE-2025-55559, CVE-2025-55560
[CVE ID]
CVE-2025-55551
[Description]
An issue in the component torch.linalg.lu of pytorch v2.8.0 allows attackers to cause a Denial of Service (DoS) when performing a slice operation.
[Additional Information]
This issue was labeled as high priority by pytorch community.
[VulnerabilityType Other]
Type Error
[Vendor of Product]
https://github.com/pytorch/pytorch
[Affected Product Code Base]
pytorch - <=2.8.0
[Affected Component]
a PyTorch API: `torch.linalg.lu`
PyTorch compiler: Inductor (https://github.com/pytorch/pytorch/tree/main/torch/_inductor)
[Attack Type]
Remote
[Impact Denial of Service]
true
[Attack Vectors]
The hacker requested to compile the PyTorch model consisting of `torch.linalg.lu` with inductor, which can result in a system crash, causing Denial of Service.
[Reference]
https://github.com/pytorch/pytorch/issues/151401
[CVE ID]
CVE-2025-55552
[Description]
pytorch v2.8.0 was discovered to display unexpected behavior when the components torch.rot90 and torch.randn_like are used together.
[Additional Information]
This issue was labeled as high priority by the PyTorch community.
[VulnerabilityType Other]
Incorrect calculation
[Vendor of Product]
https://github.com/pytorch/pytorch
[Affected Product Code Base]
pytorch - version - <=2.8.0
[Affected Component]
a combination of PyTorch APIs: `torch.rot90` and `torch.randn_like`
PyTorch compiler: Inductor (https://github.com/pytorch/pytorch/tree/main/torch/_inductor)
[Attack Type]
Remote
[CVE Impact Other]
Silent Incorrectness
[Attack Vectors]
The hacker requested to compile the pytorch model consisting of `torch.rot90` and `torch.randn_like` with inductor, which can result in incorrect output results.
[Reference]
https://github.com/pytorch/pytorch/issues/147847
[CVE ID]
CVE-2025-55553
[Description]
A syntax error in the component proxy_tensor.py of pytorch v2.7.0 allows attackers to cause a Denial of Service (DoS).
[Additional Information]
This issue was labeled as high priority by PyTorch community. This vulnerability is fixed in the latest PyTorch version (https://github.com/pytorch/pytorch/pull/154645)
[VulnerabilityType Other]
Syntax Error
[Vendor of Product]
https://github.com/pytorch/pytorch
[Affected Product Code Base]
pytorch - <=2.7.0
[Affected Component]
a PyTorch API: `torch.Tensor.random_()`
PyTorch compiler: Inductor (https://github.com/pytorch/pytorch/tree/main/torch/_inductor)
[Attack Type]
Remote
[Impact Denial of Service]
true
[Attack Vectors]
The hacker requested to compile the PyTorch model consisting of `torch.Tensor.random_()` with inductor, which can result in a Syntax Error, causing Denial of Service.
[Reference]
https://github.com/pytorch/pytorch/issues/151432
https://github.com/pytorch/pytorch/pull/154645
[CVE ID]
CVE-2025-55554
[Description]
pytorch v2.8.0 was discovered to contain an integer overflow in the component torch.nan_to_num-.long().
[Vulnerability Type]
Integer Overflow
[Vendor of Product]
https://github.com/pytorch/pytorch
[Affected Product Code Base]
pytorch - <=2.8.0
[Affected Component]
a combination of PyTorch APIs: `torch.nan_to_num` and `.long()`
PyTorch compiler: Inductor (https://github.com/pytorch/pytorch/tree/main/torch/_inductor)
[Attack Type]
Remote
[CVE Impact Other]
Silent Incorrectness
[Attack Vectors]
The hacker requested to compile the PyTorch model consisting of `torch.nan_to_num` and `.long()` with inductor, which can result in Integer Overflow when meeting `inf` input, causing Incorrect calculation.
[Reference]
https://github.com/pytorch/pytorch/issues/151510
[CVE ID]
CVE-2025-55556
[Description]
TensorFlow v2.18.0 was discovered to output random results when compiling Embedding, leading to unexpected behavior in the application.
[Additional Information]
TensorFlow community has confirmed this vulnerability.
[Vulnerability Type Other]
Incorrect calculation
[Vendor of Product]
https://github.com/tensorflow/tensorflow
[Affected Product Code Base]
TensorFlow - version - <=2.18.0
[Affected Component]
a TensorFlow API: `tf.keras.layers.Embedding`
TensorFlow compiler: XLA (https://github.com/tensorflow/tensorflow/tree/master/tensorflow/compiler/tf2xla)
[Attack Type]
Remote
[CVE Impact Other]
Silent Incorrectness
[Attack Vectors]
The hacker requested to compile the TensorFlow model consisting of `tf.keras.layers.Embedding` with XLA, which can result in silent incorrectness, causing the TensorFlow model to make wrong or dangerous decisions.
[Reference]
https://github.com/tensorflow/tensorflow/issues/82317
[CVE ID]
CVE-2025-55557
[Description]
A Name Error occurs in pytorch v2.7.0 when a PyTorch model consists of torch.cummin and is compiled by Inductor, leading to a Denial of Service (DoS).
[Additional Information]
This issue was labeled as high priority by pytorch community. This vulnerability is fixed in the latest pytorch version (https://github.com/pytorch/pytorch/pull/151931)
[Vulnerability Type Other]
Name Error
[Vendor of Product]
https://github.com/pytorch/pytorch
[Affected Product Code Base]
pytorch - <=2.7.0
[Affected Component]
a PyTorch API: `torch.cummin`
PyTorch compiler: Inductor (https://github.com/pytorch/pytorch/tree/main/torch/_inductor)
[Attack Type]
Remote
[Impact Denial of Service]
true
[Attack Vectors]
The hacker requested to compile the PyTorch model consisting of `torch.cummin` with inductor, which can result in Name Error, causing Denial of Service.
[Reference]
https://github.com/pytorch/pytorch/issues/151738
https://github.com/pytorch/pytorch/pull/151931
[CVE ID]
CVE-2025-55558
[Description]
A buffer overflow occurs in pytorch v2.7.0 when a PyTorch model consists of torch.nn.Conv2d, torch.nn.functional.hardshrink, and torch.Tensor.view-torch.mv() and is compiled by Inductor, leading to a Denial of Service (DoS).
[Additional Information]
This vulnerability is fixed in the latest pytorch version (https://github.com/pytorch/pytorch/pull/151887)
[Vulnerability Type]
Buffer Overflow
[Vendor of Product]
https://github.com/pytorch/pytorch
[Affected Product Code Base]
pytorch - <=2.7.0
[Affected Component]
a combination of PyTorch APIs: `torch.nn.Conv2d`, `torch.nn.functional.hardshrink` and `torch.Tensor.view-torch.mv()`
PyTorch compiler: Inductor (https://github.com/pytorch/pytorch/tree/main/torch/_inductor)
[Attack Type]
Remote
[Impact Denial of Service]
true
[Attack Vectors]
The hacker requested to compile the PyTorch model consisting of `torch.nn.Conv2d`, `torch.nn.functional.hardshrink`, and `torch.Tensor.view-torch.mv()` with inductor, which can result in Buffer Overflow, causing Denial of Service.
[Reference]
https://github.com/pytorch/pytorch/issues/151523
https://github.com/pytorch/pytorch/pull/151887
[CVE ID]
CVE-2025-55559
[Description]
An issue was discovered TensorFlow v2.18.0. A Denial of Service (DoS) occurs when padding is set to 'valid' in tf.keras.layers.Conv2D.
[Additional Information]
TensorFlow community has confirmed this vulnerability.
[Vulnerability Type Other]
Runtime Error
[Vendor of Product]
https://github.com/tensorflow/tensorflow
[Affected Product Code Base]
TensorFlow - version - <=2.18.0
[Affected Component]
a TensorFlow API: `tf.keras.layers.Conv2D`
TensorFlow compiler: XLA (https://github.com/tensorflow/tensorflow/tree/master/tensorflow/compiler/tf2xla)
[Attack Type]
Remote
[Impact Denial of Service]
true
[Attack Vectors]
The hacker requested to change the parameter `padding='valid'` of `tf.keras.layers.Conv2D`, which is compiled by XLA. In such a case, `tf.keras.layers.Conv2D` will throw a runtime error because it receives a negative dimension size.
[Reference]
https://github.com/tensorflow/tensorflow/issues/84205
[CVE ID]
CVE-2025-55560
[Description]
An issue in pytorch v2.7.0 can lead to a Denial of Service (DoS) when a PyTorch model consists of torch.Tensor.to_sparse() and torch.Tensor.to_dense() and is compiled by Inductor.
[Additional Information]
This issue was labeled as high priority by pytorch community. This vulnerability is fixed in the latest pytorch version (https://github.com/pytorch/pytorch/pull/151897)
[Vulnerability Type]
Not Implemented Error
[Vendor of Product]
https://github.com/pytorch/pytorch
[Affected Product Code Base]
pytorch - <=2.7.0
[Affected Component]
a combination of PyTorch APIs: `torch.Tensor.to_sparse()` and `torch.Tensor.to_dense()`
PyTorch compiler: Inductor (https://github.com/pytorch/pytorch/tree/main/torch/_inductor)
[Attack Type]
Remote
[Impact Denial of Service]
true
[Attack Vectors]
The hacker requested to compile the PyTorch model consisting of `torch.Tensor.to_sparse()` and `torch.Tensor.to_dense()` with inductor, which can result in Not Implemented Error, causing Denial of Service.
[Reference]
https://github.com/pytorch/pytorch/issues/151522
https://github.com/pytorch/pytorch/pull/151897
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment