CVE-2026-73678

Name of the Vulnerable Software and Affected Versions MindsDB Minds Platform versions prior to 26.1.1

Description An unauthenticated remote code execution flaw allows attackers to execute arbitrary OS commands. The issue occurs when crafted prompts are submitted to the unprotected 'POST /api/v1/responses/' endpoint, which interacts with the Anton agent's scratchpad tool. This tool uses the exec() function to run attacker-influenced Python source code without sandboxing. An attacker can first use the unauthenticated 'PUT /api/v1/settings/' endpoint to configure a custom LLM API key and then send a prompt to trigger the scratchpad tool. This results in full OS command execution with the privileges of the user running the desktop application, potentially exposing SSH keys, stored credentials, and environment secrets.

Recommendations Update MindsDB Minds Platform to a version later than 26.1.0. Restrict access to the 'POST /api/v1/responses/' and 'PUT /api/v1/settings/' endpoints to prevent unauthenticated access.