The vulnerability in LaVague 0.2.35 occurs in the PythonFromMarkdownExtractor.extract_as_object function, which evaluates Python code that originates from language model output derived from web page content. This improper execution of untrusted code (CWE-94) allows an attacker to run arbitrary Python on the host that hosts LaVague, leading to complete compromise of confidentiality, integrity, and availability.

The affected product is LaVague 0.2.35, developed by lavague-ai. Users running this specific version face the risk. No other versions are listed as affected, but the vulnerability description references the same function across the code base.

The CVSS score of 9.2 indicates critical severity, and although an EPSS score is not available, the lack of a patch and the open-source nature heighten the risk. Attackers can exploit the flaw by crafting malicious web pages that reference the extractor, using indirect prompt injection to insert Python code that will be executed on the operator’s host without any review. The vulnerability is remote and exploitable by anyone able to influence the content processed by the extractor.