Internal AI touches your most sensitive knowledge. Before you connect a single document, work through this checklist.
The fastest way to lose trust in an internal AI tool is for it to surface something a person should never have seen. Trust is not a feature buried in a settings page. It is the foundation, and it has to be designed in from the start.Here is the checklist we would run before connecting any AI system to company documents.
1. Access control
The single most important question: can the system ever answer from a document the user is not allowed to open?Does access inherit from your existing repositories, or do you rebuild permissions by hand?If an employee cannot open a file in SharePoint, is it guaranteed the AI will not use it to answer them?When permissions change, do the AI answers change with them?The right answer: answers come only from documents a user already has the right to see, and access inherits automatically.
No answers from documents the user cannot access. No public model training. No employee surveillance.
2. Privacy, training, and analytics
Data residency. Where does your data live? It should stay inside your own subscription boundary, with nothing copied out of your control.Model training. Your content should never train a public or shared model. Full stop.Analytics. If management gets usage insight, it should be aggregated and anonymized. Patterns, never individuals. The tool should not be able to monitor or score people.
- Answers only from documents the user can already access.
- Your data never trains a public model, and nothing leaves your boundary.
- Management sees patterns, never individuals. SOC 2 Type II in progress, with private-deployment options.
The short version
Safe internal AI comes down to four promises: it answers only from documents the user can already access, your data never trains a public model, management never sees individuals, and nothing leaves your boundary. If a tool cannot make all four, it does not belong near your company knowledge.
I look forward to seeing how these developments will improve service levels and customer satisfaction in the freight industry!