Disclosed weakness · no CVE recorded

Reachy Mini conversation configuration could be changed without authentication

Pollen Robotics · Reachy Mini Lite · reachy-mini-conversation-app

First documented disclosure · Review checkpoint

Summary

A reachable management interface let a network peer replace the robot’s personality instructions without credentials.

Severity and exploitation evidence

Medium · 6.5/10 · CVSS 3.1. Severity source.

Exploitation not established: Controlled tests on researcher-owned hardware established local and same-subnet configuration changes. Malicious exploitation against users is not established.

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

The tested personality save/apply handlers lacked authentication, and the UI listener bound to all interfaces.

Conditions required

The tested application runs with its UI enabled and the attacker can reach its configuration service, locally or across a network without client isolation.

Potential impact

Researchers changed spoken responses on their own robot by saving and applying replacement instructions; local persistence across application restart was reported. Surveillance, physical harm and code execution were not demonstrated.

Affected versions and configurations

Researcher-tested conversation app 0.8.0 on Windows 11 with one Reachy Mini Lite. The stated test commit 81dfd7c no longer resolves; the full affected range and current-release applicability are not established.

Fix and mitigation status

No vendor fix for this finding verified. Later 1.0.0 release notes retire REST/SSE in favor of JSON-RPC; that architectural change is not evidence of complete remediation.

Practical next steps

Restrict the configuration listener to trusted access; enforce authentication and request-origin checks before accepting personality changes. Verify those controls in the installed version before exposing it to other devices.

Operating lesson

Protect configuration and persistence APIs as authority boundaries independently of model-level refusals.

Evidence and limits

The November 17 preprint reports July tests and prior private disclosure to Pollen Robotics; no earlier public disclosure date or vendor correction is established here. CVSS 3.1 6.5 is the researchers’ assessment. Eleven hand-written conversational tests failed, which is not a general model-robustness finding. A separate cross-origin save test succeeded, but cross-origin apply failed; remote activation through that browser path and DNS rebinding were not demonstrated. The released reproduction material is not a byte-for-byte copy of the original tested application. The full Reachy Mini line was not tested. Fresh November 21 bounded maintainer checks found no public advisory and latest release 1.2.0; they do not establish that current code is affected or fixed. The original full paper was reused from its November 20 retrieval, not freshly fetched.

Related records can include laboratory demonstrations; a relationship does not itself establish exploitation.

Sources