CVE-2026-58659 Lightning‑AI CVE debrief

Who should care

Users of PyTorch Lightning versions up to 2.6.5, operators, platform administrators, vulnerability management teams, and security teams should be aware of this vulnerability and take immediate action to protect themselves. This vulnerability has a high CVSS score of 8.4, indicating a high severity level. Affected deployments should be identified, and owners should be assigned for follow-up.

Technical summary

The vulnerability exists in the _load_state function of PyTorch Lightning, which imports and executes module names from checkpoint _instantiator hyperparameters. An attacker can create malicious checkpoint files that bypass the weights_only=True protection, allowing for arbitrary code execution when LightningModule.load_from_checkpoint is called. This vulnerability has a high CVSS score of 8.4, indicating a high severity level. The affected product is PyTorch Lightning versions up to 2.6.5.

Defensive priority

High priority should be given to updating PyTorch Lightning to a version beyond 2.6.5, as this vulnerability allows for remote code execution. Compensating controls should be implemented to monitor and restrict access to checkpoint files.

Recommended defensive actions

• Update PyTorch Lightning to a version beyond 2.6.5
• Implement compensating controls to monitor and restrict access to checkpoint files
• Conduct regular inventory checks to ensure all instances are updated
• Monitor for suspicious activity related to checkpoint file loading
• Review relevant monitoring, detection, and logs for exposed assets that need extra review
• Track exceptions, retest remediated assets, and close the item only after evidence is documented
• Confirm whether affected product deployments exist in managed environments and assign an owner for follow-up

Evidence notes

The CVE record was published on 2026-07-15T18:16:48.743Z and was last modified on 2026-07-18T02:17:10.147Z. The NVD entry is currently Awaiting Analysis. There is limited information available about the vulnerability, and defenders should verify the affected scope and severity. The CVE record was created based on limited source details, and additional information may be needed to fully understand the vulnerability.

Sources and references

Methodology and review provenance

AI-assisted synthesis based on stored public vulnerability evidence. System validation, approval state, and publication status do not by themselves establish human review of this revision. PatchSiren helps prioritize defensive review and does not prove exposure or remediation on any system.