The vulnerability exists in the dmlc DGL library, versions up to and including 2.1.0. The function load_info/_read_torch_data in utils.py incorrectly accepts a path argument that is later deserialized without proper validation. This deserialization flaw allows an attacker to supply crafted data via the path parameter, leading to arbitrary code execution. The flaw is triggered remotely, meaning an unauthenticated attacker can send a request that forces the library to deserialize malicious input.

Affected systems are installations of dmlc DGL where the library is loaded and the _read_torch_data function is invoked. All versions up to 2.1.0 are vulnerable; newer releases may have remedied the issue, but the CVE does not list supported versions beyond that. The vulnerability applies to any platform where DGL runs, including Python environments on servers, workstations, or containers.

The CVSS score of 5.1 indicates moderate severity, but the existence of a public exploit and the remote nature of the attack suggest a higher real‑world risk. EPSS is not available, and the vulnerability is not currently listed in CISA KEV, but the lack of an official patch combined with the public exploit makes it prudent to treat it as a high‑priority risk. Administrators should assess whether their deployments use vulnerable versions and adopt the remediation steps outlined below.