CVE-2026-132122

Agentic Radar could send an OpenAI key to a scanned repository’s chosen endpoint

splx-ai · Agentic Radar

First documented disclosure · Review checkpoint

Summary

An untrusted repository’s environment file could redirect authenticated model requests.

Severity and exploitation evidence

Medium · 5.7/10 · CVSS 4.0. Severity source.

Exploitation not established: Malicious exploitation remains unverified in the reviewed sources.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

Automatic dotenv discovery loaded configuration from a repository or parent directory into the scanner process.

Conditions required

The environment-file search reaches attacker-controlled content, an OpenAI key is available, and the scan or prompt-hardening path calls OpenAI.

Potential impact

The user’s OpenAI API key can be sent to another host.

Affected versions and configurations

0.7.0 through versions before 0.14.2.

Fix and mitigation status

0.14.2.

Practical next steps

Upgrade to 0.14.2 or later. Load environment files only with an explicit trusted --env-file path or exported configuration, and rotate any exposed keys.

Operating lesson

Do not inherit service endpoints from the repository being analyzed.

Evidence and limits

Original public disclosure 2026-11-12; CVE publication 2026-11-18. The security-relevant public pull request opened November 12; the CNA uses November 13 as its public date. Version 0.14.2 shipped November 13 and includes the correction.

Sources