Skip to main content
Live Webinar with SANS (June 25)— Agentic CTI Automation for Fun & ProfitRegister Free
Mallory
Critical

RCE in NVIDIA NeMo model loading via malicious .nemo metadata

IdentifiersCVE-2025-23304CWE-94· Improper Control of Generation of…

CVE-2025-23304 is a high-severity vulnerability in NVIDIA NeMo’s model loading component. NeMo .nemo and .qnemo model files are TAR-based archives that include model metadata (for example, model_config.yaml) alongside model weights. Prior to the fix, NeMo passed metadata from these model files into Meta Hydra’s hydra.utils.instantiate() without sufficient validation or sanitization. Because Hydra instantiate() can invoke arbitrary Python callables rather than only safe class constructors, an attacker can craft model metadata so that loading a malicious .nemo file causes execution of attacker-controlled code, including via dangerous targets such as builtins.exec, builtins.eval, or os.system. NVIDIA addressed the issue in NeMo 2.3.2 by adding configuration validation and an allow list of approved packages and expected imports. Successful exploitation may result in remote code execution and data tampering.

Share:
For your environment

Are you exposed to this one?

Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.

ANALYST BRIEF

Impact, mitigation & remediation

What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.

Impact

What an attacker gets, and what they’ve been doing with it.

Successful exploitation allows arbitrary code execution in the security context of the process loading the malicious NeMo model. This can enable full compromise of the host or runtime environment used for model loading, including execution of arbitrary system commands, modification or tampering of model artifacts and local data, installation of additional payloads, theft of accessible secrets or data, and follow-on lateral movement depending on the privileges and environment of the affected process.

Mitigation

If you can’t patch tonight, do this now.

Do not load .nemo or .qnemo model files from untrusted sources. Treat model metadata as untrusted input and restrict model ingestion to trusted, integrity-verified artifacts only. Where possible, isolate model loading in a sandboxed or low-privilege environment, restrict outbound network access from model-loading workflows, and implement artifact provenance and signing checks for models obtained from repositories such as Hugging Face. If immediate upgrade is not possible, review and harden any code paths that pass untrusted metadata into Hydra instantiate()-style mechanisms.

Remediation

Patch, then assume compromise.

Upgrade NVIDIA NeMo to version 2.3.2 or later. NVIDIA’s fix adds configuration validation before execution, including an allow list of approved packages from NeMo, PyTorch, and related libraries, and verification that imports match expected classes and modules. Replace or revalidate previously obtained .nemo/.qnemo model files from untrusted or unverifiable sources before reuse.
PUBLIC EXPLOITS

Exploits

No public exploits tracked yet. Mallory keeps watching.

VALID 0 / 0 TOTALView more in app

No public exploit code observed for this vulnerability.

EXPOSURE SURFACE

Affected products & vendors

Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.

VendorProductType
NvidiaNemoapplication

Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.

ACTIVITY FEED

Recent activity

8 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.

What this page doesn’t show

The version that knows your environment.

This page is what’s public. Mallory adds the parts that aren’t: which of your assets are affected, which adversaries are exploiting it right now, which detections to deploy, and what to do tonight.
Exposure mapping

Query your assets running an affected version, and investigate the blast radius.

Threat actor evidence

Every observed campaign linking this CVE to a named adversary.

Associated malware

Malware families riding this exploit, with evidence and IOCs.

Detection signatures

YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.

Vendor-by-vendor mapping

Cross-references every affected SKU, including bundled OEM variants.

Social activity4

Community discussion across Reddit, Mastodon, and other social sources.