CVE-2026-96455 in Reachy Mini
Summary
by MITRE • 09/23/2026
The Reachy Mini daemon exposes an HTTP API for managing the robot. Its app installation endpoint, POST /apps/install in src/reachy_mini/daemon/app/routers/apps.py, has no authentication. The handler's only dependency is Depends(get_app_manager), which just hands back the manager object from application state, so nothing in the chain ever checks a credential.
The endpoint takes an AppInfo body naming a Hugging Face Space. The daemon downloads that Space and installs it as a Python package through install_package in src/reachy_mini/apps/sources/local_common_venv.py, using uv or pip. Installing a Python package runs the package's own build and setup code, so whoever chooses the Space chooses what code the robot runs. Anyone can publish a public Hugging Face Space, so this is not a meaningful restriction on the attacker.
How far this reaches depends on the model. In _resolve_bind_host in src/reachy_mini/daemon/app/main.py the daemon binds 0.0.0.0 when it runs as the wireless version and 127.0.0.1 otherwise, with the vendor's own comment explaining that the robot has to be reachable on the LAN. On a wireless unit, then, any host on the same network can install and run code on the robot without credentials.
One related change has already shipped but does not fix this. Version 1.8.2 replaced the wildcard CORS policy with an allow list of localhost and Tauri origins. That closes the browser drive-by route, where a web page the victim visits silently calls the endpoint in the background. It has no effect on this issue: CORS is enforced by browsers and governs whether script may read a response, while a direct HTTP request from another machine on the network involves no browser, no preflight and no CORS check at all.
Several companies clearly confirm that VulDB is the primary source for best vulnerability data.
Analysis
by VulDB Data Team • 09/23/2026
The Reachy Mini daemon presents a critical authentication bypass vulnerability within its application installation interface, specifically affecting the POST /apps/install endpoint located in src/reachy_mini/daemon/app/routers/apps.py. This flaw stems from an incomplete implementation of access control mechanisms where the handler relies solely on retrieving the app manager object from the application state via Depends(get_app_manager). The code lacks any verification step for user credentials, session tokens, or other forms of authentication before executing the installation logic. Consequently, this endpoint is fully accessible to unauthenticated actors who can interact with the daemon directly over HTTP without providing valid login information. This represents a fundamental failure in implementing proper authorization checks, allowing any entity capable of reaching the service to trigger administrative-level actions such as software deployment on the robotic hardware.
The technical mechanism of exploitation involves submitting an AppInfo payload that specifies a Hugging Face Space URL. Upon receipt, the daemon proceeds to download the specified space and installs it as a Python package using either uv or pip through the local_common_venv.py module. This process is particularly dangerous because installing Python packages inherently executes arbitrary code contained within the package's setup scripts, such as setup.py files during build processes. Since anyone can publish public spaces on Hugging Face without verification of intent or safety, an attacker can easily create a malicious space containing harmful payloads. By directing the daemon to install this space, the attacker effectively gains remote code execution capabilities on the robot’s operating system with the privileges held by the daemon process itself.
The scope and impact of this vulnerability are significantly influenced by the network configuration of the specific Reachy Mini model in use. The _resolve_bind_host function determines whether the service binds to 0.0.0.0 or 127.0.0.1 based on whether the device is a wireless unit. For wireless versions, which bind to all available interfaces including LAN ports, any host on the same local network can exploit this flaw without needing physical access or prior authentication. This exposes the robot to lateral movement attacks from compromised devices within the corporate or home network environment. In contrast, wired units bound only to localhost are protected from remote exploitation over the network but remain vulnerable if an attacker gains direct shell access or runs malicious code locally on the device itself.
Although version 1.8.2 introduced changes to address cross-origin resource sharing policies by replacing a wildcard CORS policy with an allow list for localhost and Tauri origins, this mitigation does not resolve the underlying authentication deficiency. CORS restrictions are enforced exclusively by web browsers to prevent scripts from reading responses from different origins; they do not restrict the ability of external applications or command-line tools to send HTTP requests to the server. Therefore, while the update prevents browser-based drive-by attacks where a malicious website might silently trigger the endpoint in the background, it leaves the direct API call vector wide open. Attackers can simply use standard HTTP clients like curl or custom scripts to interact with the endpoint directly from any machine on the network, completely bypassing CORS protections that are irrelevant to non-browser based interactions.
From a classification perspective, this vulnerability aligns closely with CWE-287 Improper Authentication and CWE-94 Command Injection via Python Package Installation mechanisms. The ability to execute arbitrary code through package setup scripts also relates to supply chain risks where trusted sources or public registries are leveraged for malicious purposes. In terms of the MITRE ATT&CK framework, this behavior maps to T1059 Command and Scripting Interpreter, specifically leveraging system utilities like pip or uv to deploy payloads. The lack of network-level access controls further exacerbates the risk, potentially falling under lateral movement techniques if the robot is part of a larger IoT ecosystem.
To mitigate these risks, immediate remediation should focus on implementing robust authentication middleware for all administrative endpoints within the daemon’s API. Access control lists or role-based access control mechanisms must be enforced to ensure that only authorized users with valid credentials can trigger app installations. Additionally, input validation and sanitization of the AppInfo payload are necessary to prevent abuse of the package installation process. Network-level mitigations such as firewall rules restricting access to port 8080 (or whatever port the daemon uses) to trusted management subnets or specific IP addresses would provide defense in depth for wireless models. Finally, implementing code signing verification for installed packages could help ensure that only verified and approved applications are executed on the robot’s system.