Executive brief
ExecuTorch is Meta's framework for deploying machine learning models on edge devices. A heap buffer overflow vulnerability in the model loading process can be exploited remotely without authentication to achieve arbitrary code execution, compromising any device running affected ExecuTorch versions. This threatens the confidentiality, integrity, and availability of systems using this framework.
Technical details
A heap-based buffer overflow exists in ExecuTorch's model loading mechanism, specifically in the prim_ops::et_copy_index() function. The vulnerability arises from insufficient bounds checking when processing model buffers during deserialization. An attacker can exploit this by crafting a malicious ExecuTorch model file and delivering it to a target system; no authentication or user interaction is required. The attack is network-reachable, allowing remote exploitation. Successful exploitation enables arbitrary code execution with the privileges of the ExecuTorch process, potentially leading to complete system compromise. The fix is available in commit ede82493dae6d2d43f8c424e7be4721abe5242be, with patched versions released as 0.7.0 across all package ecosystems (pip, Maven, Swift).
Affected products
- Meta Platforms, Inc ExecuTorch < 0.7.0
- PyTorch executorch (pip) < 0.7.0
- PyTorch executorch-android (Maven) < 0.7.0
- PyTorch executorch (Swift) < 0.7.0
Timeline
- 2025-08-07: disclosed: Published by NVD
- 2025-08-08: advisory: Published to GitHub Advisory Database as GHSA-9m39-3mf3-xwch
- 2025-08-07: patched: Fix available in commit ede82493dae6d2d43f8c424e7be4721abe5242be, version 0.7.0