Critical RCE Vulnerabilities Discovered in Major AI/ML Libraries

On January 13, 2026, researchers from Unit 42 uncovered critical remote code execution (RCE) vulnerabilities in major open-source AI/ML Python libraries utilized by companies including Apple, Salesforce, and NVIDIA. These vulnerabilities allow attackers to exploit flaws by loading model files embedded with malicious metadata, potentially leading to arbitrary code execution. Notably, NVIDIA has issued CVE-2025-23304, while Salesforce has released a patch for CVE-2026-22584, both rated high in severity. Organizations leveraging these libraries must prioritize patching to mitigate risks associated with these exploitable vulnerabilities, as the implications stretch beyond individual companies, threatening broader ecosystem security. The urgency for robust security practices in the AI/ML domain is more critical than ever, as reliance on these technologies increases across industries. Learn more.