We know that introducing AI into your intranet raises important questions around security, privacy, and accuracy. Here are some of the most common ones we hear.
Q: Is our data secure?
Yes. Aria Search is powered by Microsoft Azure OpenAI and runs in a secure, enterprise grade environment, within the same geographical region as your Oak data.
Your data is not used to train public AI models, and it is not shared outside of your organisation. Content is processed only to generate responses for your users, and remains within Oak’s trusted infrastructure and Microsoft’s controlled AI services.
Q: Is our content used to train the AI?
No. Your internal content is not used to train or improve any underlying AI models.
Each query is handled in isolation and responses are generated only from the content you have explicitly approved for Aria Search.
Q: Can Aria Search make things up?
Aria Search is designed to reduce hallucinations by grounding responses entirely in your approved internal content only.
If an answer cannot be confidently generated from that content, Aria will not invent information or pull from external sources. This helps ensure employees receive accurate, trustworthy answers they can rely on.
Q: Who can control what Aria Search uses?
Admins stay in full control.
You decide which areas are approved for Aria Search, and you can remove areas at any time. Removing an area simply stops that content from being used by AI. It does not delete anything or change how the content is managed elsewhere in Oak.
Q: Is this compliant with enterprise and regulated environments?
Aria Search is built with enterprise use in mind. Using Microsoft Azure OpenAI provides strong foundations for security, data handling, and responsible AI practices, making it suitable for information heavy and regulated organisations.
Q: What country is the data Aria uses hosted in?
All AI services used by Oak are deployed within the same region as the tenant's data centre. The reason AI responses contain American spelling is due to the data used to train the underlying models, as primarily American, being a US-owned operation (OpenAI/Azure).