Path Traversal in Edge0-AI Audio8_TTS

Executive Summary

This vulnerability is a path traversal flaw in Edge0-AI Audio8_TTS's ONNX runtime (onnx_runtime_0_1b_int8/arktts_runtime). The runtime reads ONNX file paths from runtime_manifest.json without validating them, allowing attackers to specify paths outside the model package (e.g., '../victim/slow_selected.onnx'). This can lead to silent model substitution, where an attacker's malicious ONNX graph is loaded instead of the intended one.

Useful 0 Not Useful 0

Detection Guidance

Check for suspicious ONNX model files in runtime_manifest.json that reference paths outside the model directory, such as those containing '..' or absolute paths. Inspect the model directory for unexpected files like slow_selected.onnx or codec_internal.onnx in unauthorized locations. Use file integrity tools to compare current model files against known-good versions.

Useful 0 Not Useful 0

Impact Analysis

An attacker could exploit this to replace the intended model with a malicious one, potentially gaining control over generated speech content (integrity), disrupting the runtime (availability), or leaking private data (confidentiality). If custom op libraries are enabled, arbitrary code execution may be possible.

Useful 0 Not Useful 0

Compliance Impact

This vulnerability could lead to unauthorized access to sensitive data (e.g., speech content), violating GDPR's data protection principles and HIPAA's confidentiality requirements. Non-compliance risks include legal penalties, reputational damage, and loss of trust.

Useful 0 Not Useful 0

Mitigation Strategies

Validate all ONNX file paths in runtime_manifest.json to ensure they resolve within the model directory. Reject any paths containing '..' or absolute paths. Update the runtime to enforce strict path containment checks. Monitor for unauthorized model substitutions or tampering attempts.

Useful 0 Not Useful 0