CVE-2026-7680: COCO Annotator Path Traversal Vulnerability

CVE-2026-7680 Overview

CVE-2026-7680 is a path traversal vulnerability in jsbroks COCO Annotator versions up to 0.11.1. The flaw resides in the backend/webserver/api/datasets.py file within the Data Endpoint component. Attackers can manipulate the folder argument to traverse outside the intended directory structure. The vulnerability is remotely exploitable and requires only low-privileged authentication. A public exploit has been released, increasing the risk of opportunistic attacks. The vendor was contacted prior to public disclosure but did not respond. This issue is tracked under CWE-22 (Improper Limitation of a Pathname to a Restricted Directory).

Critical Impact

Authenticated remote attackers can read files outside the intended dataset directory by manipulating the folder parameter, potentially exposing sensitive data on the host system.

Affected Products

• jsbroks COCO Annotator versions up to and including 0.11.1
• Component: Data Endpoint (backend/webserver/api/datasets.py)
• Deployments exposing the COCO Annotator web service to authenticated users

Discovery Timeline

• 2026-05-03 - CVE-2026-7680 published to NVD
• 2026-05-05 - Last updated in NVD database

Technical Details for CVE-2026-7680

Vulnerability Analysis

The vulnerability exists in the dataset management API of COCO Annotator, an open-source image annotation tool used for computer vision and machine learning workflows. The folder parameter passed to the dataset endpoint in backend/webserver/api/datasets.py is not properly sanitized before being concatenated into a filesystem path. An authenticated attacker can supply traversal sequences such as ../ to escape the intended dataset root directory. This allows access to arbitrary directories on the underlying host filesystem accessible to the application process. The flaw is reachable over the network and a working exploit has been published in a third-party research repository.

Root Cause

The root cause is improper input validation on the folder argument within the Data Endpoint handler. The application trusts user-supplied path components without canonicalizing them or enforcing a base-directory boundary check. This pattern matches CWE-22, where pathname construction fails to restrict access to a designated directory.

Attack Vector

Exploitation requires network access to the COCO Annotator web interface and a valid low-privileged account. The attacker submits a crafted request containing directory traversal sequences in the folder parameter. The backend resolves the manipulated path and returns directory contents or file metadata from outside the dataset root. Public exploit details are available in the Thoropass Vulnerability Research Repository and VulDB Entry #360833.

No verified code examples are available. Refer to the VulDB advisory and the linked research repository for proof-of-concept details.

Detection Methods for CVE-2026-7680

Indicators of Compromise

• HTTP requests to dataset API endpoints containing ../, ..\, or URL-encoded traversal sequences (%2e%2e%2f) in the folder parameter.
• Application logs showing dataset folder resolution to paths outside the configured dataset root directory.
• Unexpected file or directory enumeration responses returned by backend/webserver/api/datasets.py.

Detection Strategies

• Inspect web server and application logs for requests targeting the datasets API with anomalous folder values.
• Deploy web application firewall rules to flag traversal patterns in query strings and JSON bodies sent to COCO Annotator endpoints.
• Correlate authenticated session activity with filesystem access events outside the dataset directory.

Monitoring Recommendations

• Monitor read access on sensitive host paths from the COCO Annotator process user.
• Alert on requests where the folder argument contains path separators or encoded traversal payloads.
• Track the rate of dataset API calls per authenticated user to detect enumeration behavior.

How to Mitigate CVE-2026-7680

Immediate Actions Required

• Restrict network access to COCO Annotator deployments using firewall rules or reverse-proxy allowlists until a fix is available.
• Audit existing user accounts and disable any low-privileged accounts that are not actively required.
• Review dataset directory contents and host filesystem layout to identify sensitive files reachable by the service account.

Patch Information

No official vendor patch is available. The vendor was contacted prior to disclosure but did not respond. Operators should monitor the jsbroks COCO Annotator GitHub repository for future updates and consider applying community fixes that canonicalize the folder parameter against the dataset base directory.

Workarounds

• Run COCO Annotator inside an isolated container with read-only mounts limited to the dataset directory.
• Apply a reverse-proxy rule that rejects requests containing .., %2e%2e, or absolute path indicators in the folder parameter.
• Enforce least-privilege filesystem permissions on the service account so traversal cannot reach sensitive system files.
• Disable public network exposure and require VPN or zero-trust access to the annotation interface.

# Example NGINX rule to block traversal payloads in the folder parameter
location /api/dataset/ {
    if ($args ~* "folder=.*(\.\.|%2e%2e|/etc/|/root/)") {
        return 403;
    }
    proxy_pass http://coco_annotator_backend;
}