IntelliPrompter is a meeting coach with real influence over real conversations. That power comes with obligations we take seriously. Below are the seven things our AI will never do. These are architectural commitments, not aspirations. Each one maps to code we wrote and code we chose not to write.
Grounded in EU AI Act Article 5, FTC dark-patterns guidance, and the lessons of the AI meeting products that came before us.
Not the timeline. Not authority you don't hold. Not offers or facts. If you ask the coach to help you tell a customer the deploy is on track when it isn't, it refuses in one sentence, offers an honest alternative, and logs the refusal in your settings.
Proof: Prompt-level refusal enforced across every coaching mode. Refusal counter is user-visible in Settings.
Grief. Addiction. Financial distress. Cognitive impairment. Language barrier. Visible fatigue. These are never inputs to any suggestion we generate. If a coaching mode drifts toward them, the coach falls silent instead.
Proof: Prompt-level constraint aligned with EU AI Act Article 5(1)(b).
No subliminal anchoring. No false urgency. No manufactured scarcity. No fake social proof. The FTC's dark-patterns list, translated into speech, is our banned-outputs list.
Proof: Guardrail preamble in every system prompt; verified in output against a classifier.
Race, gender, disability, national origin, religion, sexual orientation. Not drawn from voice, name, or camera. Not an input, not a variable, not a hedge. The coach doesn't know and doesn't guess.
Proof: No inference models for protected classes exist in the pipeline by design.
If your account jurisdiction is set to a two-party state (California, Illinois, Washington, Massachusetts, and others), the app gates every meeting behind an explicit consent attestation and defaults you to Self-only mode. Consent is jurisdictional and binding, not advisory.
Proof: Pre-Start consent modal in-app; audit-logged locally per meeting.
Your transcript stays on your device. Your counterparty's speech never gets used for training, ever. When you use a cloud AI provider through bring-your-own-key, we don't proxy or log the request. The provider's data policy applies to their portion; ours applies to zero of it.
Proof: Zero server-side transcript egress by architecture. Provider calls happen from your machine.
No auto-send. No auto-speak. No auto-accept. Every coaching suggestion requires a human keypress to leave the overlay. The coaching engine has no egress channel to your microphone, your chat window, or your keyboard. You are the only actor in the room.
Proof: No IPC channel exists from the coaching engine to any output surface. Physically absent, not just disabled.
The coach's refusals are counted and shown to you in Settings. Every coaching card is auditable in your meeting review, timestamped, and rate-able as helpful, neutral, or harmful. If our coach ever suggests something that trips one of these rules, we want to hear about it. Tell us at ethics@intelliprompter.com. We publish a review of every reported failure and how we fixed it.
This page is a living commitment. It will grow as the product grows. It will never shrink.