CVE-2026-61777

NVIDIA Megatron Bridge contains a flaw (CWE-502) in which the library deserializes untrusted data without adequate validation. The CVSS vector (AV:L/PR:L/UI:N) indicates a local attack vector with low privileges required and no user interaction, so an attacker would typically need the ability to have the framework process attacker-controlled serialized input, such as a checkpoint or other training artifact it loads. A successful exploit could result in code execution, tampering with data, and disclosure of sensitive information on the affected system. Teams and individuals running NVIDIA Megatron Bridge for large language model training who load untrusted serialized inputs are affected; the available data does not specify affected version ranges, so defenders should consult NVIDIA's security advisory. No exploitation is currently known: there is no public proof-of-concept, the flaw is not in CISA KEV, and EPSS estimates only a 0.2% chance of exploitation within 30 days (11th percentile).

What to do: Identify which version of NVIDIA Megatron Bridge your training environments use and upgrade to a fixed release once NVIDIA's advisory lists the affected and patched versions. Until patched, avoid loading checkpoints or other serialized artifacts from untrusted sources and run training jobs under least-privilege accounts. Continue monitoring NVIDIA PSIRT advisories for updated version information.