Auditing of AI Assist

Hi,

We're hoping to being making use AI Assist in our deployment of new Redwood pages but first we need to satisfy some stakeholders, particularly around the auditing of it.

I've reviewed the AI Data Security Principles and just would like some clarifications.

• When the AI Assist button is clicked, is the input and response recorded anywhere for us to review at a future date?

• I note that metrics are recorded in OCI Object Storage, what kind of metrics does this include and is that also accessible to us?

• Is there any data available to us around the performance of the Guardrail service? For example, the amount of requests rejected (jail-breaking attempts? etc).