This vulnerability in MLflow (<=3.10.1.dev0) involves missing resource-level authorization checks on multipart upload endpoints under the --serve-artifacts mode. Attackers can exploit this to overwrite artifacts of other users, leading to unauthorized writes and the risk of model supply chain poisoning. The compromised models may allow arbitrary code execution upon loading. The vulnerability is fixed starting from version 3.10.0. The CVSS v3 base score is 9.0, indicating critical severity. No known exploits are reported in the wild. The vendor advisory from Red Hat confirms the issue and remediation version.

Successful exploitation allows attackers with limited privileges to overwrite artifacts belonging to other users, enabling unauthorized cross-user writes. This can poison the model supply chain and potentially lead to arbitrary code execution when the compromised models are loaded, posing a critical risk to system integrity and confidentiality.

Upgrade MLflow to version 3.10.0 or later, where this vulnerability is fixed. Since the vendor advisory confirms the fix in 3.10.0, applying this official update is the recommended remediation. No alternative mitigations or temporary fixes are indicated in the advisory.