Principles

These are intentions, stated so they can be checked. They are not evidence that every property already meets them.

What we are betting on

Once sufficiently capable AI exists, being smarter than other people stops being the scarce resource. Judgment, values, incentives, and deciding what intelligence should be pointed at become the bottleneck. This is a thesis, not a settled fact, and it may be wrong.

What we are trying to favor

Scale is treated as a means, not the goal.

How decisions are meant to be checked

The design has three peers with reciprocal controls: IntelliPulse (sensing), Galatea (reasoning and constraint) and Skighnet (execution). The sequence describes a workflow, not a chain of command, and each is meant to be able to reject or reopen an action. No single model or person is meant to be an unrestricted authority. Independent enforcement by Galatea is not yet established; see status.

How this could fail

Challenge us

If you think a principle is hollow, a claim is unsupported or a failure mode is missing, tell us.