NVIDIA Megatron Bridge: deserialization of untrusted data allows code execution in the training job

CVE-2026-61754AI/ML frameworks & servingcurated

Impact

Deserialization of untrusted data in Megatron Bridge can lead to code execution, data tampering, and information disclosure inside a training job. The blast radius is the job's own context on the GPU node - GPUs, mounted training data, output checkpoints, and job credentials. What makes this fleet-relevant rather than single-host is artifact flow: checkpoints and conversion inputs routinely cross team and tenant lines, and each one is an execution vector into whatever job consumes it. NVIDIA fixed eight of these together and gave no per-CVE input detail.

Who can reach it

Local, low privileges required, no user interaction - an attacker able to place or supply a serialized artifact that a Megatron Bridge process loads.

What to do

Upgrade Megatron Bridge to 0.5.1; versions 0.0 through 0.5.0 are affected. Rebuild training images against the fixed package and restart jobs. No node drain, reboot, or firmware maintenance required.

References

https://nvd.nist.gov/vuln/detail/CVE-2026-61754 https://github.com/NVIDIA/product-security/tree/main/2026/5868 https://www.cve.org/CVERecord?id=CVE-2026-61754

Related entries

All AI/ML frameworks & serving entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.