High-Severity NVIDIA NeMo Framework Flaws Allow Code Injection and Privilege Escalation in AI Pipelines

NVIDIA has released a security update for its NeMo Framework on 14 November 2025, addressing two high-severity vulnerabilities that could enable code execution, privilege escalation, information disclosure, and data tampering in AI pipelines. The first flaw, CVE-2025-23361, affects a NeMo script that inadequately validates attacker-controlled input, while the second, CVE-2025-33178, targets the BERT services component where malicious data can trigger unsafe code injection.

Both vulnerabilities carry a CVSS base score of 7.8, indicating a high-severity local attack vector and are described as relating to code generation control and unsafe injections, respectively. NVIDIA states that affected versions are all prior to 2.5.0, with 2.5.0 being the updated release, and notes the impact across Linux, Windows, cloud container images, and custom infrastructure built using NeMo.

Patches are available from GitHub and PyPI, and the advisory emphasises the potential for a successful exploit to lead to code execution, privilege escalation, information disclosure, and data tampering.