We take the responsibility of using Large Language Models (commonly referred to as "AI") in your family communications very seriously. Here's exactly how we approach it.
Hush uses IBM Granite, the world's first major LLM model family to pass ISO 42001 certification — the international standard that assesses LLM model ethics, accountability, and trust.
ISO 42001 is an independent, rigorous audit of how an LLM system is developed, governed, and deployed. By choosing a certified model, we're choosing one that has been held to a verifiable, external standard and trained on ethical data sources — not just relying on a company's own claims about safety and training data.
We never use LLM agents — systems where an LLM takes actions autonomously, makes decisions on your behalf, or operates without your direct oversight.
In Hush, the LLM plays a specific, bounded role: with your particular co-parenting relationship as context, it reads and rewrites text. That's it. It does not decide who receives messages, when they are sent, or what actions are taken. Every consequential decision stays with you.
When the LLM rewrites an outgoing message, you see a preview before anything is sent. You choose to send it or try again with a different strategy — the LLM's version is only sent when you decide to send it.
When the LLM filters an incoming message, the original is always available for you to read. We never hide what was sent to you — we only offer a calmer version first. Transparency is not optional.
Hush is clear about what the LLM has done. Filtered messages are shown as filtered. LLM-generated previews are shown as previews. We don't obscure the role the LLM plays — we make it visible so you can make informed decisions.
Your communications with your co-parent involve your children's wellbeing. We take that seriously. Our goal is to give you a tool that reduces conflict, while keeping you fully informed and fully in charge.