The AI policy shouldn't arrive as a consultant's document. Almost all of it already exists in things Echidna has ratified: your 2022 Principles, your leadership diagnostic answers, and the red lines from July 29. This pre-read shows the mapping, so on September 1 we're confirming and filling gaps rather than starting from a blank page.
Each block below pairs a line from the Echidna Giving Principles (the Board-ratified 2022 document, now reproduced on the workspace) with the AI rule that follows from it. Where the diagnostic or the July 29 session already produced the same rule, the sources agree, and that's the point: the policy holds because it's yours.
We read and react to materials that are shared with us in order to learn and to acknowledge the time and effort involved in keeping us informed.
Staff personally read grantee materials before AI touches them. AI may then organize thinking, check an assessment for gaps, or help format. This settles the scenario 3 question from the diagnostic at the principle level: reading grantee submissions is how Echidna honors grantee effort.
We aim to contribute without being controlling: we let grantees take the lead. We try to uplift our grantees' voices and hand the microphone over to them.
AI never substitutes for direct interaction with grantees, and grantee-facing writing keeps the author's own voice. AI can help prepare for a conversation; it doesn't have the conversation.
We avoid bureaucracy and we try not to create unnecessary speed bumps. We value efficiency over perfection.
The policy stays short enough to actually be used: a one-page core, a red/yellow/green data guide, and account rules. For internal research the principles already bless directional AI use, since you look for directional information rather than publishable evidence.
We don't innovate for the sake of innovation.
AI gets adopted task by task where it adds value, driven by the team's own wish list from the diagnostic. No tool or feature enters the workflow just because it exists. This is also, nearly word for word, one of your own diagnostic principles: used only where it adds value.
We assume that everyone is acting with good intentions. Our response to failure is measured: what matters is learning from failure, not avoiding failure altogether.
Blameless incident reporting. If someone puts the wrong thing in the wrong tool, the expectation is to say so immediately, and the response is to fix and learn. A policy people are afraid of is a policy that never hears about incidents.
We are guided by the ideas, knowledge, and expertise of those closest to the problems. All employees are comfortable and encouraged to express opinions, share ideas, and influence decisions.
Staff shape the tool decision through the diagnostic, and the policy gets reviewed with the whole team before it's final. Concerns raised (accuracy, privacy, environment) get addressed in the document rather than around it.
We believe in the value of diversity, equity, and inclusion. We recognize and try to correct for our biases as an organization based in the United States.
Training includes a module on AI bias, with specific attention to how models represent the marginalized communities Echidna serves. Outputs about grantees and communities get the same bias check you already apply to your own work.
One leadership diagnostic response closed with five principles that read as a near-draft preamble. Set against the 2022 document, they're the same values in AI form:
"Only as a way to enhance judgment and learning, not in ways that circumvent it. Never in replacement for human interaction with grantees, colleagues, etc. Human review and judgment are required for every product produced by AI; ultimate responsibility lies with a human, and AI cannot be blamed for shortcomings. No sensitive information to be shared with public versions of AI and other necessary controls for safeguarding private information. Used only where it adds value, not just because it is there (mindful of environmental costs)."
From the leadership diagnostic and confirmed or refined in the July 29 session (full notes here):
Read the mapping above with one question in mind: is there anywhere you think the AI rule doesn't follow from the principle? Those are the spots worth our discussion time on September 1.
Draft prepared by SA Advisory, August 2026, as the pre-read for the September 1 policy working session. Please shout if anything looks off. Full timeline on the schedule page.