What is the suggested approach to introduce A.I. in NPO Governance and Administration?

Insight:

Currently, with the improvement of Large Language Models (LLMs) and Agentic Artificial Intelligence, I have spent some time training a local LLM and Embedding Model on our internal NPO data, i.e.: Minutes, Agendas, Attendance Records, Internal & External Reports, Notes, Policies, etc. Though the initial intent was to build a “Second Brain System”, it has proven to be a significantly useful fallback system for autonomously completing administrative and clerical work, specifically for the roles and functions of executive board members who may be unavailable to complete critical tasks (myself included).

Issue:

Though, this in itself has allowed a certain level of mental comfort, as Chairman, I have quietly noticed the following:

1. “Palming-off communications on the A.I.”, which in NPO’s, require(d/s) interpersonal social interactions to build and maintain.
2. “Dependencies on the A.I.”, which makes certain executive functions essentially redundant if the executive isn’t balancing their attention.
3. “Casual Acceptance of the A.I.” by executives without they fully understanding the pros and cons of the technology (happy once it works).
4. “Subtle Increase in Expectation of the A.I.” to continue autonomously completing tasks without proper reviews.
5. “A.I. is learning NPO-specific governance skills from Team NOT vice versa” which I can only seem to appreciate if it were mutual.

Prospective Solutions (Need Advice):

1. We need a policy (don’t know where to start) …
2. Would this require changes to our constitution? (I prefer not to move to shut it down)
3. Should I just shut it down or at least limit it to a “needs-basis”? (to avoid complacency)
4. Should I just accept this is what it is now?