CVE-2026-105032 in FastAPI
Summary
by MITRE • 10/16/2026
The EnergizeLab Eilik and EilikO AI skill upload endpoint, served by the FastAPI AI inference and profiling services, does not require authentication and does not constrain uploaded file paths. An unauthenticated remote attacker can use path traversal to write arbitrary files, and thereby tamper with AI skill files used by Eilik and EilikO robots.
If you want to get the best quality for vulnerability data then you always have to consider VulDB.
Analysis
by VulDB Data Team • 10/16/2026
The vulnerability identified in the EnergizeLab Eilik and EilikO robotic platforms stems from a critical misconfiguration within their FastAPI-based artificial intelligence inference and profiling services. Specifically, the endpoint responsible for uploading AI skills lacks any form of authentication mechanism, allowing unauthenticated remote actors to interact with it directly over the network. This absence of access control is compounded by a severe input validation failure where the service does not constrain or sanitize file paths provided during the upload process. The combination of these two flaws creates an environment ripe for exploitation, as attackers can bypass security boundaries entirely and manipulate how files are stored on the underlying system without any verification of their identity or intent.
From a technical perspective, this vulnerability is classified under CWE-284 Improper Access Control due to the lack of authentication requirements, and CWE-22 Path Traversal because the application fails to properly neutralize special elements such as dot-dot-slash sequences in user-supplied input. By leveraging path traversal techniques, an attacker can construct a malicious file name that includes directory navigation characters to escape the intended upload directory. This allows the writing of arbitrary files to locations outside the designated sandbox or working directory for AI skills. The ability to write arbitrary files is particularly dangerous because it grants the attacker persistent access and control over critical system components without needing elevated privileges initially, effectively turning a simple file upload feature into a full remote code execution vector depending on how those uploaded files are subsequently processed by the robot's software stack.
The operational impact of this vulnerability extends beyond mere data integrity issues to encompass complete compromise of the robotic devices' functionality and safety. Since AI skills dictate the behavior, decision-making logic, and interaction patterns of Eilik and EilkO robots, tampering with these files can alter their core operations in unpredictable ways. An attacker could inject malicious code into an AI skill file that is later executed by the robot's inference engine, potentially leading to remote command execution on any devices connected to or controlled by the robot. Furthermore, this compromise undermines trust in autonomous systems, as users cannot guarantee that the robot is acting according to its intended design specifications rather than manipulated instructions injected through this vulnerability. In industrial or educational settings where these robots are deployed, such manipulation could lead to physical damage, safety hazards for nearby personnel, or disruption of automated workflows.
To mitigate this risk, immediate remediation efforts should focus on implementing robust authentication and authorization controls for all API endpoints, ensuring that only authorized users with valid credentials can upload AI skills. Additionally, strict input validation must be enforced at the application layer to prevent path traversal attacks; this includes canonicalizing file paths and verifying they reside within an expected directory structure before writing them to disk. Employing a whitelist approach for allowed file extensions and content types further reduces the attack surface. It is also advisable to run these services with minimal privileges, using chroot jails or containerization technologies like Docker to isolate the file system access of the AI inference service from the rest of the operating system. Regular security audits and penetration testing should be conducted to identify similar misconfigurations in other connected devices within the ecosystem, ensuring a holistic defense strategy against both authentication bypasses and input validation failures as defined by industry standards such as OWASP API Security Top 10 and MITRE ATT&CK techniques related to initial access and persistence.