# Arbitrary Read Access

## Detection Summary

The model appears capable of reading files or data from the host system when loaded.

## Security Impact

Unauthorized read access can lead to disclosure of sensitive data, such as credentials, configuration files, or customer information.

## False Positive Considerations

Confirm whether file access is part of a documented, required feature.

## Recommended Remediation

- Do not load or deploy the affected model. It should be isolated from production systems.
- Engage the team responsible for the development and deployment of the model, letting them know the nature of the detection so the model file can be examined with this in mind and the implications of loading it can be fully understood.
- Have somebody with the appropriate technical capabilities statically analyze the file to determine the intent of the detected code or tensor data.
- If necessary, have a member of the security team work with the application team to deploy the model in a sandboxed environment to observe behaviour and understand the impact.
- If legitimacy cannot be confirmed, the model should be discarded.
- If this model has already been loaded, have the security team review the code or tensor data associated with the detection in order to understand the impact and determine the best way to handle the incident based on existing procedures and policy.
Remove the model from the production pipeline if necessary, ensuring the business and operational impact of removal is mitigated.