MindsDB Minds Platform: Unauthenticated Remote Code Execution
MindsDB Minds Platform up to 26.1.0 exposes unprotected API endpoints that allow unauthenticated attackers to execute arbitrary OS commands with the privileges of the MindsDB process. CVSS 10.0.
- Vendor
- MindsDB
- Product
- Minds Platform
- CVSS
- 10.0
- EPSS (exploit probability)
- 0.8%
- Status
- patched
- Published
MindsDB’s Minds Platform (all versions through 26.1.0) exposes two API endpoints without any authentication requirement. The POST /api/v1/responses/ endpoint routes requests to the Anton agent’s scratchpad tool, which executes attacker-supplied Python source via exec() without sandboxing. A separate PUT /api/v1/settings/ endpoint — also unauthenticated — lets any caller configure the LLM API key the instance uses, enabling an attacker to pre-position before triggering code execution.
The result is arbitrary OS command execution as whatever user runs the MindsDB process. No credentials, no prior access, no user interaction required.
Affected versions: all versions through 26.1.0
What to do: Update to a fixed release. If you cannot patch immediately, firewall the MindsDB API port — this service should not be internet-accessible under any circumstances. Audit credentials and data stored in or accessible to any exposed instance.
See the full write-up for action items and priority guidance.
