Copilot for Microsoft 365 Security and Governance AMA

I think AI provides great opportunities for everyone, but what I'd like to know is: 1. Can Microsoft provide any insight into the rough percentage of time Copilot has a hallucination? 2. What kind of "success rate" is the minimum standard Microsoft provides with Copilot? (For example, in only 3% of tests it hallucinates, so a 97% success rate) 3. if Copilot has a hallucination while being used and provides wrong, inaccurate feedback (such as for a stupid simple example recommends we disable firewall). How will we engage Microsoft of a dangerous and incorrect recommendation? Thank you