Study finds unauthenticated configuration attacks on LLM-powered Reachy Mini Lite robot
Researchers report that a network-exposed management interface on the Reachy Mini Lite lets attackers alter robot personality and system prompts without authentication, bypassing conversational defences.
In plain words
A robot that uses a large language model can be changed not through what you say to it, but through its setup interface. Researchers say an attacker on the same network could rewrite the robot's instructions without logging in, changing how it behaves. The robot's ability to resist trick-filled conversations did not stop this.
What was announced
In an arXiv paper, researchers describe a configuration-layer attack against the Pollen Robotics Reachy Mini Lite, an embodied robot that relies on a cloud LLM. An unauthenticated attacker could use the management interface to replace personality settings and the cloud model's system prompt, with no adversarial conversational input required. The authors report demonstrating both local and network-adjacent access, and both configuration attack paths produced observable behavioural changes on physical hardware. By contrast, eleven prompt-injection attempts across five categories were reported as resisted, which the authors say shows that inference-layer refusal cannot prevent unauthorised changes to model instructions.
The paper maps the weakness to CWE-306 (missing authentication for a critical function) and CWE-1327 (binding to an unrestricted IP address), with a reported CVSS v3.1 score of 6.5, Medium. A configuration endpoint also reportedly accepts cross-origin requests without a CORS preflight (CWE-352, also 6.5 Medium), and the persistent-memory sink stores content without enforcing its stated restrictions. Potential consequences named by the authors include surveillance through onboard cameras and microphones and social engineering of nearby users. A review of 59 public repositories found missing authentication in 10 of 11 source-verifiable applications, or 11 of 11 counting a framework default; the authors state this indicates recurrence in the sample, not prevalence in commercial products.
What it does not show
- The evidence is the paper's abstract only; full methods, device details and vendor response are not covered by the supplied passage.
- Findings rest on the authors' own demonstrations and repository sample; no independent confirmation is included.
- The repository survey covers open-source projects and, per the authors, does not establish prevalence in commercial products.
For security and IT teams
- Scope
- The reported issues concern the Pollen Robotics Reachy Mini Lite's network-exposed configuration interface and a sample of 59 public repositories, not a broad product audit.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.