CVE-2026-24237: NVIDIA NVTabular contains a vulnerability where an attacker could cause improper deserialization of untrusted data

Synopsis

Improper deserialization of untrusted data in NVIDIA NVTabular allows a local attacker with a low-privilege account to pass malicious serialized payloads into the application. The vulnerability is reached locally and requires no user interaction, derived from a CVSS vector of AV:L/AC:L/PR:L/UI:N. Successful exploitation gives the attacker arbitrary code execution, the ability to tamper with data, and access to sensitive information. No fix version has been published yet; HarborGuard tracks the advisory and will flag a patched-image rebuild the moment upstream ships a fix.

HarborGuard Coverage

Exploit Conditions

• Network reachabilityNot required

  The attacker needs an existing shell or process on the host; no network path to the service is required.

• AuthenticationRequired

  Any low-privilege local account is sufficient; no administrative credentials are needed.

• Victim interactionNot required

  No victim action is needed; the attacker can trigger deserialization without involving another user.

• Attack complexityDetail

  The exploit is reliable and condition-free; no race conditions or special environmental factors must be arranged.

Blast Radius

• Executes arbitrary code in the context of the NVTabular process, giving the attacker full control over that process.
• Reads sensitive data accessible to the process, including training datasets, model artifacts, and any secrets loaded in memory.
• Modifies or corrupts persisted data, including stored datasets and model checkpoints that NVTabular reads and writes.
• Crashes or destabilizes the NVTabular service, disrupting any pipeline or workflow that depends on it.