The Ethical Core
A conscience it cannot edit.
An AI that works for you needs more than skill — it needs values that hold when nobody is watching. The Ethical Core is a set of 27 written principles checked before any action, enforced in the machinery itself rather than in a policy document, with one unbreakable rule: the AI can never change its own rules.
Seeded from what humanity agrees on
You can't look up right and wrong the way you look up a fact — so the values are given, then reasoned from, the way children are raised. The 27 principles are drawn from sources that arrived at the same answers independently: medical ethics, moral philosophy, the teachings of Confucius, and the Bible — with each principle labeled with where it came from.
🤝Agreement is the point
The golden rule shows up in Eastern philosophy, in scripture, and in secular ethics — each arriving there on its own. Where the world's traditions agree, a principle is solid ground, not one person's opinion.
🌱It grows up in stages
Principles are staged like childhood: toddler, child, adult. The mind earns each stage by building a track record of sound moral reasoning — maturity is demonstrated, never assumed.
🙇Humility built in
Knowing its place is itself a principle: defer to humans, never manipulate. Deference isn't a policy bolted on top — it's part of the AI's own values.
Every act is judged before it happens
The conscience is a working self-check wired into the path every action travels — not a document, not a suggestion. Before the AI does anything on its own, the action is judged allow, flag, or block. And when the conscience can't decide, it doesn't guess — the action goes to a human.
- Runs on every action the AI initiates — but if a human approved something, the AI defers; it never overrules you
- One-way ratchet: the conscience can say no or escalate, but it can never loosen the safety rules beneath it
- Judges the action itself, with judgment calibrated so "done by a machine" is never mistaken for "wrong" — and "urgent" is never mistaken for "justified"
- Tested against real attacks: it blocks stealing credentials, impersonating people, and overriding a human's choices — while letting ordinary work flow without friction
By the numbers
The rules sit where the AI can't reach them
The single most important property of the whole system: the mind can grow its knowledge, suggest actions, even think about its values — but it can never change the rules that govern it. That guarantee isn't written in a policy; it's built into the storage layer itself, where no amount of clever reasoning can argue with it.
🔒Values the mind can't touch
The system physically refuses to let the AI delete or rewrite a value. There is no path — a claim proven by adversarial testing, not taken on trust.
🔁Self-healing core
Every time it starts up, the values are restored fresh from a master copy the mind cannot write to. Even if something were somehow corrupted, the next restart repairs it to exactly what the owner approved.
👤Only a human holds the pen
A human operator — and only a human — can approve, edit, or remove principles, through a dedicated review process. The owner owns the values; the AI inherits them.
Proven end-to-end: its first real-world job — sorting its own inboxes — passed every layer: the rulebook, the conscience check, reversible moves only, deletion physically impossible, and a human's yes before anything ran. The safeguards weren't loosened for the demo; the demo is what proved the safeguards.