If "Enable AI Features" setting is enabled, then all KB articles (including tech-only articles) are sent to the OpenAI API to calculate vector embedding. This is not your standard text-based or chat-based interface, but a technical endpoint used to convert text to array of numbers (vectors).
Open AI doesn't store any information on their servers and we have a DPA-agreement signed with them. Your data is not used for training. Nothing that can associate a KB article with your company name or other "identifiable" info, unless you actually have this info in the KB article text.
The resulting vectors are stored on our side, on a EU-based server. Customer's data is isolated from each other.
Also, no automated responses are being sent to a ticket-creator unless a technician decides to proceed. There is no risk of leaking any private information without an explicit consent from one of your employees. The content of a support ticket is only sent to AI if you actually decide to generate an AI response. Nothing happens automatically (unless you set up an automation rule "Generate AI response" of course)
We do the same for any external web pages that you decide to index. We send them to OpenAI for generating additional context information.