The vulnerability in langchain-openai before version 1.1.14 involves a TOCTOU race condition in the _url_to_size() helper function used for image token counting. This function validates URLs to prevent SSRF but performs the validation and the actual fetch in separate network operations with independent DNS resolutions. An attacker can exploit this by controlling a hostname that resolves to a public IP during validation and then to a private or localhost IP during the fetch, bypassing SSRF protections. This is classified as CWE-918 (Server-Side Request Forgery). The CVSS 3.1 base score is 3.1, reflecting low impact with network attack vector, high attack complexity, no privileges required, and user interaction required. No official remediation or patch is currently documented.

The impact is limited to potential SSRF attacks that could allow an attacker to make the server perform unauthorized requests to internal or localhost network resources. The CVSS score of 3.1 indicates low confidentiality impact and no integrity or availability impact. There are no known exploits in the wild, and the vulnerability requires user interaction and has high attack complexity.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Since the vulnerability affects versions prior to 1.1.14, upgrading to version 1.1.14 or later (if available) is recommended once confirmed. Until an official fix is released, consider restricting outbound network requests from the affected component or implementing additional network-level protections to prevent SSRF exploitation.