Safety in a system of many models
Orchestrating many models means setting shared rules they all follow.
When many intelligences cooperate, safety cannot be the job of any single one. It has to live in the layer that coordinates them.
- Bounded permissions: tools and connectors only get the access a task needs.
- Human confirmation before anything is sent, bought or deleted.
- Clear notices in regulated areas like health, law and engineering.
- Your data is yours: export or delete everything at any time.
These principles come from the NeoAI Constitution and apply to every model Neo uses.
