onnx — Vulnerabilities & Security Advisories 13

ONNX serves as an open format for machine learning models, enabling interoperability across frameworks. Historically, vulnerabilities have included remote code execution, buffer overflows, and improper input validation, often stemming from insecure parsing of model files. While no major public incidents have been widely documented, the 9 recorded CVEs highlight potential risks in model processing and serialization. Security characteristics include reliance on third-party runtime environments and dependencies, which may introduce additional attack surfaces. Proper validation of untrusted models remains critical to prevent exploitation, as malformed inputs could lead to arbitrary code execution or system compromise.