CVE-2023-43472: 
NixOS vulnerability analysis and mitigation

An issue in MLFlow versions 2.8.1 and before allows a remote attacker to obtain sensitive information via a crafted request to REST API. The vulnerability was discovered in September 2023 and affects the MLflow framework, a popular platform for managing machine learning lifecycles (Contrast Security). The vulnerability received a CVSS v3.1 base score of 7.5 HIGH (NVD).

The vulnerability exists in the MLflow user interface's REST API, which fails to properly validate content-type headers. While POST requests normally use application/JSON content type that triggers a preflight request, the API's failure to check content type headers allows requests with text/plain content type to bypass these checks. This misconfiguration enables attackers to make simple requests to localhost without triggering a preflight check (Contrast Security).

The vulnerability poses a significant threat to ML models and their training data. A successful exploit could lead to the exfiltration of both the serialized version of the ML model and all training data. This represents an existential threat to companies built around ML models, equivalent to gaining write access to a software company's source code. Additionally, there's potential for model poisoning, where attackers could inject malicious data into the model's training pool (Contrast Security).

Users of MLflow are strongly advised to upgrade to the latest version as soon as it becomes available. The vulnerability was reported to Databricks, the original creator and current maintainer of MLflow, on September 5, 2023, with a fix scheduled for release (Contrast Security).