CVE-2025-32434
Published: 18 April 2025
Summary
CVE-2025-32434 is a critical-severity Deserialization of Untrusted Data (CWE-502) vulnerability in Linuxfoundation Pytorch. Its CVSS base score is 9.3 (Critical).
Operationally, exploitation aligns with the MITRE ATT&CK technique Python (T1059.006); ranked in the top 37.0% of CVEs by exploit likelihood; it is not currently listed in the CISA KEV catalog.
This vulnerability is AI-related — categorised as Deep Learning Frameworks; in the Supply Chain and Deployment risk domain.
Deeper analysis
PyTorch versions 2.5.1 and earlier contain a remote command execution vulnerability in the model-loading path. The flaw is triggered when torch.load is called with weights_only=True, allowing deserialization of untrusted data that can execute arbitrary commands on the host. The affected component is the core tensor and autograd machinery used for GPU-accelerated neural-network workloads, and the issue is tracked as CWE-502.
An attacker who can supply a malicious model file can exploit the weakness over the network without authentication or user interaction. Successful exploitation yields full control over the victim process, including the ability to read, modify, or delete data and to pivot within the environment. Because PyTorch is commonly used to load models from external sources or shared checkpoints, the attack surface includes any application or pipeline that ingests untrusted .pt files under the weights_only flag.
The GitHub Security Advisory GHSA-53q9-r3pm-6pq6 and the accompanying Debian LTS notice state that the vulnerability is fixed in PyTorch 2.6.0; users are advised to upgrade immediately and to avoid loading untrusted models until the patch is applied. The EPSS score rose from a low baseline to a peak of 0.0122, indicating that exploitation interest increased after public disclosure.
EU & UK References
- 🇪🇺 ENISA EUVD: EUVD-2025-11856
Vulnerability details
PyTorch is a Python package that provides tensor computation with strong GPU acceleration and deep neural networks built on a tape-based autograd system. In version 2.5.1 and prior, a Remote Command Execution (RCE) vulnerability exists in PyTorch when loading a…
more
model using torch.load with weights_only=True. This issue has been patched in version 2.6.0.
- CWE(s)
AI Security AnalysisAI
- AI Category
- Deep Learning Frameworks
- Risk Domain
- Supply Chain and Deployment
- OWASP Top 10 for LLMs 2025
- None mapped
- Classification Reason
- Matched keywords: pytorch
Related Threats
MITRE ATT&CK Enterprise TechniquesAI
Why these techniques?
The RCE vulnerability in PyTorch's torch.load with weights_only=True enables arbitrary command execution via malicious models, facilitating Python interpreter abuse (T1059.006) and exploitation for client execution (T1203).
Affected Assets
Mitigating Controls
Likely Mitigating Controls AI
Per-CVE control mapping for this CVE has not run yet; the list below is derived from the weakness types (CWEs) cited in the NVD entry.
Penetration testing supplies malicious serialized objects, detecting unsafe deserialization and supporting corrective actions.
Evaluation of untrusted data handling (deserialization testing) reveals unsafe processing, which the required remediation process addresses.
Untrusted serialized data can be deserialized and observed inside the chamber, blocking gadget-chain exploitation outside the sandbox.
Validates or rejects untrusted serialized data before deserialization occurs.
Identifies and blocks malicious code introduced through deserialization of untrusted data at system boundaries.
Integrity verification of serialized information can detect tampering before deserialization occurs.
Provenance of associated data allows detection of untrusted sources before deserialization or processing occurs.