The vulnerability arises from deserialization of untrusted data in the LanguageModel class of Flair, enabling arbitrary code execution when a malicious model file is loaded. This flaw is a classic insecure deserialization weakness (CWE‑502) that can compromise confidentiality, integrity, and availability if an attacker can control model input. The impact is the ability to run arbitrary code with the privileges of the process that loads the model.

Flair, a machine learning library, is affected in all releases starting with 0.4.1 through the most recent version. Any deployment that loads models using the LanguageModel class is at risk. The vulnerability is listed against the Flair product across all affected releases.

The CVSS score of 8.4 indicates high severity. The EPSS score is below 1 %, suggesting limited current public exploitation activity, and it is not currently listed in the CISA KEV catalog. However, the vulnerability allows remote code execution, forcing users to treat it as high risk. Attackers would typically supply a crafted model file or supply a maliciously encoded model through any ingestion pathway. The exploitation requires only the ability to trigger the loading of a model file, so environments that allow model uploads or downloads are especially vulnerable.