A critical deserialization vulnerability in NVIDIA NeMo Speech can lead to remote code execution and full system compromise.

The vulnerability exists in the TabularTokenizer class, which performs deserialization of untrusted, attacker-controlled .pkl files using pickle.load() without validation. This is an unauthenticated, remotely exploitable flaw categorized as CWE-502.

With a CVSS score of 8.8, this vulnerability poses a severe threat to any application utilizing the affected versions of NVIDIA NeMo Speech. Exploitation can result in arbitrary code execution, data tampering, and information disclosure, potentially leading to complete loss of confidentiality, integrity, and availability of the host system.

Immediate Action: Apply vendor security updates immediately as provided by NVIDIA.

Proactive Monitoring: Monitor network and application logs for unusual file handling or attempts to process external pickle files.

Compensating Controls: Implement strict input validation and ensure that the application only processes serialized objects from trusted, verified sources.

Public Exploit Available: No

The reliance on insecure deserialization makes this a high-priority risk. Users of NVIDIA NeMo Speech must update to the latest version once available to eliminate the vulnerability, as no other effective mitigation exists for this architectural flaw.