The vulnerability is a stored cross‑site scripting flaw in the _genai/_evals_visualization component of the Google Cloud Vertex AI Python SDK. It allows any unauthenticated remote actor to embed benign‑looking escape sequences into the model evaluation results or dataset JSON that are later rendered in a victim's Jupyter or Colab session. The embedded JavaScript can then execute in the context of the victim's environment, enabling actions such as reading cookies, tampering with page content, or exfiltrating sensitive data. The flaw is classified as a CWE‑79 XSS weakness.

Google Cloud Vertex AI SDK for Python, versions 1.98.0 through 1.130.x inclusive, is affected. The vulnerability does not apply to releases prior to 1.98.0 or to 1.131.0 and newer.

Based on the description, the attack vector is an unauthenticated attacker injecting script escape sequences into model evaluation results or dataset JSON, which is then displayed in a victim's Jupyter or Colab environment. The CVSS score of 8.6 classifies the flaw as high severity, while the EPSS score of less than 1% indicates a low current probability of exploitation. The CVE is not listed in CISA's KEV catalog. If exploited, arbitrary JavaScript execution in the victim's browser can lead to data theft, session hijacking, or manipulation of the analysis workflow.