CVE-2026-58449 is a critical code injection vulnerability in neuml's txtai product affecting versions up to 9.10.0. The /reindex API endpoint accepts a function parameter resolved via txtai.util.Resolver, which uses __import__ and getattr on user-supplied input without an allowlist. When the API is exposed without a configured TOKEN (authentication is opt-in) and the index is writable, an attacker can specify an arbitrary callable such as subprocess.getoutput, leading to remote code execution as the server process during reindexing. The vulnerability is mitigated by a fix that restricts access to the endpoint behind a new reindex configuration flag.

Successful exploitation allows unauthenticated remote attackers to execute arbitrary code on the server process running txtai, resulting in full compromise of confidentiality, integrity, and availability of the affected system. This can lead to complete system takeover, data theft, or service disruption. However, exploitation requires the API to be exposed without authentication and the index to be writable, which are not default settings.

A fix is available that restricts access to the /reindex endpoint behind a new reindex configuration flag. Users should apply this fix by updating txtai to a version including the commit 11b32da or later. Additionally, configuring authentication tokens (TOKEN) to protect the API and ensuring the index is not writable unless necessary will mitigate the risk. Patch status is not explicitly confirmed in the advisory, so users should check the vendor repository or release notes for the fixed version containing commit 11b32da.