Listen to this Post
How the CVE Works
CVE-2025-2149 affects PyTorch 2.6.0+cu124 in the `nnq_Sigmoid` function of the Quantized Sigmoid Module. The vulnerability arises from improper initialization when manipulating the `scale` and `zero_point` arguments. Attackers can exploit this flaw locally, though exploitation complexity is high due to required precision in input manipulation. The issue stems from insufficient validation of quantization parameters, leading to potential memory corruption or unintended behavior in model inference.
DailyCVE Form
Platform: PyTorch
Version: 2.6.0+cu124
Vulnerability: Improper Initialization
Severity: Problematic
Date: 06/23/2025
Prediction: Patch expected by 08/2025
What Undercode Say
import torch quant_sigmoid = torch.ao.nn.quantized.Sigmoid() Malicious scale/zero_point triggers vulnerability quant_sigmoid.scale = malicious_scale quant_sigmoid.zero_point = malicious_zero
How Exploit
- Local privilege escalation
- Crafted quantization params
- Memory corruption
Protection from this CVE
- Update PyTorch
- Validate quantization params
- Restrict local access
Impact
- Model instability
- Local DoS
- Data leakage
Sources:
Reported By: nvd.nist.gov
Extra Source Hub:
Undercode
