What is the CVSS score of CVE-2025-33248?

CVE-2025-33248 has a CVSS 3.1 base score of 7.8 (High). CVSS vector: CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H. EPSS exploitation probability: 0.1%.

Which versions of Nvidia are affected by CVE-2025-33248?

NVIDIA Megatron-LM contains a critical unsafe deserialization vulnerability that allows remote code execution when users load maliciously crafted files.

How to fix or mitigate CVE-2025-33248?

Within 24 hours: Audit all systems running Megatron-LM and restrict file loading to trusted sources only; disable the hybrid conversion script if not actively used. Within 7 days: Implement network segmentation to isolate Megatron-LM systems from production environments; enable logging and monitoring on all file access and model loading activities. Within 30 days: Establish a vendor communication protocol to track patch availability; evaluate alternative LLM frameworks if NVIDIA does not provide timely remediation; conduct security training for teams handling model files.

Nvidia CVE-2025-33248

EUVD-2025-208975 HIGH

Deserialization of Untrusted Data (CWE-502)

2026-03-24 nvidia

GHSA-wxjh-pc4c-7rjq

RCE Deserialization Information Disclosure Nvidia

7.8

CVSS 3.1 · NVD

Severity by source

NVD PRIMARY

7.8 HIGH

AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H

Primary rating from NVD · only source for this CVE.

CVSS VectorNVD

CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H

Attack Vector

Local

Attack Complexity

Low

Privileges Required

Low

User Interaction

None

Scope

Unchanged

Confidentiality

High

Integrity

High

Availability

High

Lifecycle Timeline

EUVD ID Assigned

Mar 24, 2026 - 20:31 euvd

EUVD-2025-208975

Analysis Generated

Mar 24, 2026 - 20:31 vuln.today

CVE Published

Mar 24, 2026 - 20:24 nvd

HIGH 7.8

DescriptionCVE.org

NVIDIA Megatron-LM contains a vulnerability in the hybrid conversion script where an Attacker may cause an RCE by convincing a user to load a maliciously crafted file. A successful exploit of this vulnerability may lead to code execution, escalation of privileges, information disclosure, and data tampering.

AnalysisAI

NVIDIA Megatron-LM contains a critical unsafe deserialization vulnerability (CWE-502) in its hybrid conversion script that allows remote code execution when a user loads a maliciously crafted file. The vulnerability affects NVIDIA Megatron-LM installations and enables attackers to execute arbitrary code, escalate privileges, disclose sensitive information, and tamper with data. With a CVSS score of 7.8 and local attack vector requiring low privileges and no user interaction, this represents a significant risk for organizations using this large language model training framework.

Technical ContextAI

NVIDIA Megatron-LM is a deep learning framework specifically designed for training large-scale language models using model and data parallelism techniques. The vulnerability stems from unsafe deserialization (CWE-502) in the hybrid conversion script, which processes serialized data from files without proper validation. Unsafe deserialization occurs when an application deserializes untrusted data, allowing attackers to inject malicious objects that execute arbitrary code during the deserialization process. The affected product is identified via CPE cpe:2.3:a:nvidia:megatron_lm:*:*:*:*:*:*:*:*, indicating all versions are potentially vulnerable pending specific version restrictions from the vendor. This class of vulnerability is particularly dangerous in machine learning frameworks where model files and conversion scripts regularly handle serialized data structures.

RemediationAI

Users should immediately consult the NVIDIA security advisory at https://nvidia.custhelp.com/app/answers/detail/a_id/5769 for official patching guidance, version-specific remediation steps, and validated security updates for Megatron-LM. Until patches can be applied, organizations should implement strict controls on file sources used with the hybrid conversion script, restricting processing to only trusted model files from verified origins. Additional mitigations include implementing file integrity verification before processing conversion scripts, restricting file system permissions to prevent low-privileged users from writing to directories containing model conversion files, applying principle of least privilege to limit which users can execute the conversion script, and monitoring for suspicious deserialization activity. In shared computing environments, consider sandboxing or containerization with restricted privileges when running model conversion operations to limit potential impact of successful exploitation.