CVE-2026-58659
HighCVSS 7.8Exploitation Probability (EPSS)
Low risk15th percentile - higher than 15% of all known CVEs
Summary
A vulnerability in PyTorch Lightning up to version 2.6.5 (fixed in commit d710d68) allows remote code execution via the _load_state function, which imports and executes attacker-controlled module names from _instantiator hyperparameters in checkpoint files. Attackers can craft malicious checkpoint files that bypass weights_only=True protections and execute arbitrary code when LightningModule.load_from_checkpoint is called.
Risk Assessment
The risk involves an attacker gaining control over the system by supplying a crafted checkpoint file, leading to remote code execution and potential compromise of the entire infrastructure.
Recommendation
Immediately update PyTorch Lightning to a version containing commit d710d68 or later, and avoid loading checkpoint files from untrusted sources.
Other vulnerabilities in PyTorch Lightning
See all- CVE-2026-31221High
PyTorch-Lightning versions 2.6.0 and earlier contain an insecure deserialization vulnerability in the checkpoint loading mechanism. The LightningModule.load_from_checkpoint() method calls torch.load() without setting the security-restrictive weights_only=True parameter, allowing deserialization of arbitrary Python objects.
- CVE-2026-44484Critical
A vulnerability in PyTorch Lightning versions 2.6.2 and 2.6.2 introduces functionality consistent with credential harvesting.
Original NVD description (English source)
PyTorch Lightning through 2.6.5, fixed in commit d710d68, contains a remote code execution vulnerability in the _load_state function that imports and executes attacker-controlled module names from checkpoint _instantiator hyperparameters. Attackers can craft malicious checkpoint files that bypass weights_only=True protections to execute arbitrary code when LightningModule.load_from_checkpoint is called.