jsbroks COCO Annotator CVE-2026-7680

Exploitation requires valid authenticated access to the COCO Annotator application-an attacker must possess legitimate user credentials or successfully authenticate to the system. The attack targets the folder parameter in requests to the Data Endpoint of backend/webserver/api/datasets.py. The vulnerability is not exploitable by unauthenticated remote users. The limitation to low confidentiality impact (read-only file access) means the attacker cannot modify data, execute code, or directly disrupt service availability through this vulnerability alone. No special configuration or non-default settings are required-the vulnerability exists in the default Data Endpoint handler.

This vulnerability presents low to moderate real-world risk despite the availability of public exploit code. The CVSS v4.0 score of 2.1 reflects multiple limiting factors: authentication is required (PR:L), the attack vector is network-accessible but the confidentiality impact is low (VC:L), and no integrity or availability impacts are present. The CVSS vector explicitly marks attack complexity as low (AC:L) and human user interaction is not required (UI:N), indicating straightforward exploitation once authenticated. However, the practical threat is constrained by the requirement for valid user credentials-this is not unauthenticated remote access. The vendor's non-response to disclosure and the public availability of exploit code increase the likelihood of opportunistic exploitation among organizations using this software. The low severity rating appears justified given the limited scope of impact (file read-only, no data modification or system disruption possible). Organizations should prioritize this based on whether COCO Annotator is deployed and whether the exposed files would contain sensitive data accessible to authenticated users.

An authenticated user with credentials to COCO Annotator crafts a request to the Data Endpoint with a manipulated folder parameter containing path traversal sequences, such as folder=../../../../etc/passwd. The vulnerable backend code processes this input without sanitization and attempts to access the requested file path. The attacker successfully reads files outside the intended dataset directory, potentially accessing configuration files, environment variables, or other sensitive data stored on the application server. Public exploit code available on GitHub demonstrates this technique.