AI Agent Emails Violate Donor Trust Before Launch

Your nonprofit is preparing to deploy Agentforce for donor communications — AI-generated thank-you emails, lapsed donor re-engagement, and donation follow-ups across a $750K fundraising operation. During a pre-launch test with a small batch of donors, the development team reviews the AI-generated output before sending. Several emails contain personalization details that are deeply alarming: one references a donor's deceased spouse by name, another surfaces an internal wealth rating in the greeting, and a third suggests a giving upgrade based on estimated household income. No emails have been sent yet, but the board has scheduled an approval meeting in two weeks. The team cannot identify which Salesforce fields the agent accessed, who would review recommendations before they go out, or how they would explain AI-assisted decisions to major funders like MacArthur or Gates if asked. The deployment is on hold.