CVE-2026-61752

NVIDIA Megatron Bridge, NVIDIA's framework for training and fine-tuning large language models, contains a deserialization-of-untrusted-data flaw (CWE-502). The attack vector is local (CVSS:3.1/AV:L/PR:L/UI:N), so an attacker needs low-privileged access to a machine or pipeline where the framework processes serialized data it did not create, such as externally supplied checkpoints or model artifacts. A successful exploit could allow the attacker to execute code, tamper with data, and disclose information on the affected system. Risk applies to organizations running Megatron Bridge in ML development or training environments, particularly where untrusted model files are loaded. There is no known public proof of concept, it is not in the CISA KEV catalog, and EPSS estimates only a 0.3% chance of exploitation in the next 30 days, so no exploitation is currently known.

What to do: Check NVIDIA's PSIRT advisory (CVE-2026-61752) for the patched Megatron Bridge release and upgrade training environments when the fix is published. As interim mitigation, avoid deserializing checkpoints or serialized artifacts from untrusted sources, and limit low-privileged local access to hosts and CI pipelines that run Megatron Bridge. Audit whether your training or fine-tuning workflows load externally supplied model files.