CVE-2019-6446: An issue was discovered in NumPy before 1.16.3.

Security readout for executives and security teams

Older NumPy versions loaded Python pickle data through numpy.load in a way that could run attacker-controlled code if an application opened a malicious serialized object. Business risk depends on whether systems process NumPy files from untrusted users, partners, pipelines, or storage. The CVE is disputed because this behavior can be legitimate for trusted object arrays. Exposure is most likely in Python services, notebooks, batch jobs, ML pipelines, or data-processing tools using NumPy before 1.16.3 and loading serialized arrays from untrusted or weakly authenticated sources. Prioritize remediation for internet-facing upload features, shared data platforms, and ML pipelines consuming third-party files. Lower urgency may be reasonable for isolated systems that only load trusted internal artifacts, but the trust boundary should be verified rather than assumed. Mitigation focus: Upgrade NumPy to 1.16.3 or later where vendor guidance supports it.; Do not load NumPy pickle/object data from untrusted sources.; Require authentication and provenance checks for serialized data inputs..

Prepared

Jun 7, 2026, 04:57 UTC (UTC+00:00)

Generated from the cited source records. This long-tail analysis has not been individually reviewed by a named human.