Executive brief
OneFlow is a deep learning framework used for building and scaling artificial intelligence models. A flaw in how the software handles GPU hardware requests allows an attacker to crash the application by providing an invalid or negative hardware identifier. This results in a denial-of-service, potentially disrupting AI training workloads or production inference services.
Technical details
A denial-of-service vulnerability exists in OneFlow v0.9.0 due to improper validation of GPU device indices. The flaw is located in the flow.cuda.get_device_properties() function, which fails to sanitize the device_id input before passing it to the underlying CUDA API. Specifically, providing a negative or out-of-bounds integer causes a failed assertion in cudaSetDevice() within cuda_util.cpp, leading to a core dump and application crash. An attacker capable of influencing the device index parameter in a running OneFlow environment can trigger this crash without authentication. As of the advisory, the issue was identified in version 0.9.0.
Affected products
- Oneflow-Inc OneFlow 0.9.0
Timeline
- 2025-06-30: disclosed: Issue reported on GitHub repository
- 2026-01-28: advisory: CVE published to NVD