Todb

The AI skill upload endpoint, served by the FastAPI AI inference and profiling services, lacks authentication and does not restrict uploaded file paths. This allows an unauthenticated remote attacker to perform path traversal—a method of accessing files and directories that are stored outside the web root folder—to write arbitrary files and tamper with AI skill files used by the robots.

The EnergizeLab Eilik NestJS auth service accepts a user-controlled user ID in account update requests (mass assignment). A registered user can set another account's identifier and take over that account.

Insufficient authorization checks on administrative APIs allow authenticated users with low privileges to perform unauthorized actions. This includes the ability to retrieve AI provider keys and modify system configurations and object-store contents.

The FastAPI AI inference and profiling services lack authentication for AI model administration functions. This allows an unauthenticated remote attacker to modify the AI model configuration and change the active model during runtime.

Low-privilege registered accounts can read arbitrary files from the server by providing a controlled file path. This issue allows unauthorized access to middleware files and server secrets.

EnergizeLab AI Station firmware, EilikO Linux robot firmware v2.1 20260617182218 release, and the EnergizeLab Windows desktop app 3.1.2 download firmware updates over plain HTTP and do not verify their integrity or authenticity. An attacker able to intercept or modify the traffic can substitute a malicious update.

The EnergizeLab FirmwareUpdater (AArch64 ELF, distributed in the CDN bin.zip) uses a single hard-coded AES key shared by all installations. An attacker who recovers the key from the updater can decrypt and forge firmware packages for any device that trusts it.

The EnergizeLab Eilik NestJS auth service issues device tokens for arbitrary robot serial numbers without authentication. An unauthenticated remote attacker can obtain a valid device token for any serial number and act as that device against the cloud services.

The EnergizeLab AI Station (RK3562) firmware image ships with default root credentials and SSH enabled. An attacker with network access to the device can log in as root and take full control of it.

The software contains an insecure default configuration for the LaunchURL Blueprint node `UKismetSystemLibrary::LaunchURL`. The URL scheme filter `FURLRequestFilter` is empty by default, allowing all URL schemes. On Windows, this leads to a call to the `ShellExecuteW()` function using the open verb, which enables the execution of a referenced file when a `file:///` URI is provided. Consequently, any game developed with Unreal Engine that loads user-generated content containing Blueprints is susceptible to command execution through a malicious local file or a remote file hosted on a WebDAV share.

An integer overflow occurs in the `mount volume()` function during FAT32 processing when the operation `fasize *= fs->n fats` wraps. This allows an attacker to control file-size metadata, resulting in unsafe read lengths for downstream callers. This issue can be exploited via crafted FAT images on USB sticks or SD cards to achieve arbitrary code execution on embedded devices, including those using ESP-IDF, STM32Cube, Zephyr, and ArduPilot. Additionally, remote delivery is possible through OTA or update pipelines. Real-world exploitation has been demonstrated on a camera device.

The Cowork VM image handling process validates only the presence of the file and a version marker string before booting `rootfs.img`, failing to verify the integrity of the image content at the time of use. A local attacker with unprivileged code execution as the victim macOS user can modify the VM root filesystem image to ensure it is trusted during subsequent boots. This allows for persistent arbitrary code execution within the VM and unauthorized access to host-mounted directories. This issue is characterized by a missing support for integrity check, which is a failure to verify that data has not been altered by an unauthorized party.

The embedded MQTT broker is configured to permit anonymous connections and lacks topic-level read or write Access Control Lists (ACLs). This allows any host on the same network to subscribe to sensitive telemetry topics or publish control messages directly to the robot without authentication or authorization.

The firmware contains hardcoded administrative credentials embedded in the image. These credentials are identical across all devices and cannot be modified or removed by end users, allowing unauthorized access to device management interfaces.

A hidden, persistent backdoor provides remote, unauthenticated or weakly authenticated access to privileged functionality. This backdoor is undocumented, cannot be disabled through user-facing settings, and persists after factory resets and standard firmware updates.

Name of the Vulnerable Software and Affected Versions Ollama versions up to 18.1 Description A flaw exists in Ollama up to version 18.1 related to the processing of the `server/download.go` file within the Model Pull API component. This can lead to a server-side request forgery (SSRF) attack, which can be launched remotely. The vendor was contacted but did not respond. Recommendations Versions prior to 18.2 are affected. At the moment, there is no information about a newer version that contains a fix for this vulnerability.

The software improperly processes annotation directives embedded in automatically extracted binary data, leading to arbitrary command execution when a user interacts with the user interface. The `@execute` annotation, designed for trusted, user-authored comments, is also parsed in comments generated during auto-analysis, such as CFStrings in Mach-O binaries. This allows a crafted binary to present clickable text that, when clicked, executes attacker-controlled commands on the analyst’s machine.

The encryption algorithm used to protect firmware updates is encrypted using key material accessible to attackers. This allows unauthorized modification of firmware updates, which can then be trusted by Unitree products. The issue affects the firmware generation and extraction processes. Currently, there is no publicly documented method to bypass the update process and inject malicious firmware packages without the owner's knowledge.

The HexStrike AI MCP Server is susceptible to a command injection issue. By supplying a command-line argument beginning with a semicolon (`;`) to an **API endpoint** created by the `EnhancedCommandExecutor` class, a composed command is directly executed with root privileges. The server does not sanitize these arguments in its default configuration. The `EnhancedCommandExecutor` class is the component responsible for processing commands. The `username` and `password` parameters are not explicitly mentioned as being involved in this issue.

Multiple Unitree robotic products sharing a common firmware contain a command injection issue. An attacker can inject a malicious string during WiFi configuration via a BLE module, and then trigger a restart of the WiFi service. This allows the attacker to execute commands as root through the `wpa supplicant restart.sh` shell script. The vulnerability affects devices using firmware derived from the MIT Cheetah codebase, including the G1 (humanoid) and Go2 (quadruped) branches.