CVE-2026-79785: Improper Certificate Validation in CVHub520 X-AnyLabeling

The vulnerability in X-AnyLabeling 4.0.0-beta.9 arises because the model downloader disables TLS certificate verification by using ssl._create_unverified_context() with urllib.request.urlopen. This means neither the certificate chain nor hostname is validated when fetching models over HTTPS. An attacker able to intercept the connection can supply arbitrary model files. The downloaded file is saved and then validated only by format checks (safe_check_model), without verifying hashes or signatures. For ONNX models, the attacker controls the inference output. For PyTorch (.pth or .pt) models, the file is unpickled without restrictions, allowing execution of arbitrary code on vulnerable PyTorch versions that do not default to weights_only loading.

An attacker who can intercept the model download connection can supply malicious model files that bypass format validation. This can lead to arbitrary code execution on the host running X-AnyLabeling, compromising system integrity and confidentiality. The vulnerability affects the integrity of the model loading process and can be exploited without user interaction or privileges.

No official patch or remediation is currently available. Users should avoid using version 4.0.0-beta.9 of X-AnyLabeling until a fix is released. As a temporary measure, users can manually verify model file integrity using trusted hashes or signatures if available. Network-level protections such as TLS interception detection or using secure, trusted networks may reduce risk. Monitor vendor advisories for updates.