Megatron-LM: Arbitrary code execution in the training job
CVSS 7.8CVE-2025-23264NVIDIA / GPU stackcurated
Impact
Arbitrary code execution in the training job
Who can reach it
Malicious model/checkpoint or dataset
What to do
Bump Megatron-LM in training images; rebuild; treat checkpoints as untrusted input
Fleet impact
How widespread
Common - Megatron is the reference large-model training stack for customers doing pretraining on rented clusters
Cost to remediate
hot-patch (upgrade to v0.12.1, rebuild training images)
Why it hits the whole fleet
A malicious file supplied to the Python component triggers code injection; on a shared training cluster the attacker executes inside a job that already holds cluster-wide storage and fabric credentials
References
Related entries
- Megatron-LM: Arbitrary code executionCVE-2025-23265 · Megatron-LMHigh
- Megatron-LM: Local privesc / RCE via unsafe pickle deserializationCVE-2025-33247 · Megatron-LMHigh
- Megatron-LM: RCE via insecure deserializationCVE-2025-33248 · Megatron-LMHigh
- GPU Display Driver: Local privesc via file permissionsCVE-2025-23276 · GPU Display DriverHigh
- vGPU Manager: Guest-to-host escape (stack buffer overflow)CVE-2025-23283 · vGPU ManagerHigh
- vGPU Manager: Guest-to-host escape (stack buffer overflow)CVE-2025-23284 · vGPU ManagerHigh
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.