NVIDIA Megatron-LM checkpoint flaw allows code execution
Opening a malicious checkpoint file is required; ZDI rates CVE-2026-65110 at CVSS 7.8.
Malicious checkpoint RCE in NVIDIA Megatron-LM
NVIDIA Megatron-LM has an unsafe deserialization flaw (CWE-502) in checkpoint loading. An attacker who can convince a user or job to load a maliciously crafted checkpoint can achieve remote code execution on the machine that loads it. A successful exploit may also lead to privilege escalation, information disclosure, and data tampering; CVSS 3.1 rates this high (7.8) with local access, low complexity, and low privileges. Users and operators of Megatron-LM who load checkpoints are affected; no affected version range is stated in the provided data. It is not on CISA's KEV list and no public proof-of-concept is known.
What to do: Do not load Megatron-LM checkpoints from untrusted sources or untrusted storage. Treat checkpoint files as executable input, restrict who can place files where training jobs load them, and apply NVIDIA's fix as soon as a patched release is published for your install.
| Product | Affected |
|---|---|
| NVIDIA Megatron-LM | — |
Order-of-magnitude estimate by the model from install counts, market share and public scan data it knows; verify before quoting.
NVIDIA Megatron-LM contains a vulnerability in checkpoint loading where, an attacker could cause an RCE by convincing a user to load a maliciously crafted file. A successful exploit of this vulnerability might lead to code execution, escalation of privileges, information disclosure, and data tampering.
Opening a malicious checkpoint file is required; ZDI rates CVE-2026-65110 at CVSS 7.8.
Today 1.5K readers · 2.4K views · 768 reads · all time 189K views