The ONNX machine learning framework contains a critical input validation flaw that enables unauthenticated attackers to corrupt internal object attributes, potentially leading to application instability.

The vulnerability stems from the improper use of the Python setattr function within the ExternalDataInfo class, which fails to validate keys provided in model files. This allows an unauthenticated attacker to supply a crafted ONNX model that overwrites internal object properties, leading to improper input validation and potential resource consumption issues.

The exploitation of this vulnerability poses a significant risk to the integrity and availability of systems processing machine learning models. By manipulating internal object attributes, an attacker could cause system crashes or potentially divert model execution flow, resulting in service disruption and potential compromise of data processing pipelines. With a CVSS score of 8.6, this flaw is categorized as high severity due to the lack of required authentication and the potential for significant impact on system stability.

Immediate Action: Organizations must update the ONNX package to version 1.21.0 or later to apply the necessary input validation checks.

Proactive Monitoring: Security teams should monitor for anomalous model processing behavior, such as unexpected application crashes or high CPU usage during model parsing, and review logs for errors related to attribute assignment.

Compensating Controls: Implement strict input validation or sandboxing for any machine learning models received from untrusted sources to prevent the parsing of malicious metadata.

Public Exploit Available: Unknown.

Given the high CVSS score and the potential for remote exploitation, immediate action is required to patch the affected software. Organizations should prioritize updating to version 1.21.0 to eliminate the underlying vulnerability and prevent the risk of malicious model exploitation within their production environments.