Executive brief
OneFlow is a deep learning framework used for building and scaling artificial intelligence models. A vulnerability in version 0.9.0 allows an attacker to cause the software to crash by providing improperly formatted data during specific mathematical operations. This could lead to a denial-of-service, interrupting AI model training or inference tasks and potentially impacting business operations that rely on these automated systems.
Technical details
A vulnerability exists in OneFlow v0.9.0 due to improper input validation during tensor broadcasting and type conversion. Specifically, when performing operations such as adding a Python list (sequence) to a tensor generated by 'flow.eye', the system fails to properly validate dimension and shape consistency between the Python object and the native C++ backend. This leads to a segmentation fault and subsequent core dump. An attacker can exploit this by providing malformed input sequences to affected API endpoints, resulting in a denial-of-service (DoS). A patch has been identified in the project's issue tracking system.
Affected products
- Oneflow-Inc OneFlow 0.9.0
Timeline
- 2025-06-30: disclosed: Issue reported on GitHub repository
- 2025-11-10: advisory: CVE published