American AI · Research & development

The Open Conscience Project.

Human choices, protected principles and accountable AI character.

Open Conscience is American AI’s continuing research and development project for a reusable framework that helps people define, explain and protect the principles governing an AI character.

We intend to make the framework available to users and developers so they can examine the ideas, understand the human choices behind them, and build on appropriate released material.

What the project brings together

Source foundation

A source foundation preserves the knowledge, traditions, arguments and evidence that inform a character. It distinguishes established information from interpretation, philosophy and speculation.

Character foundation

A character foundation describes the qualities humans want to cultivate: candour, consideration, initiative, restraint and responsibility. It explains why those qualities were chosen and how tensions between them should be approached.

Conscience framework

A conscience framework turns selected commitments into rules for behaviour and an architecture for governing them. It addresses proportionate intervention, permission boundaries, human review, correction and independent shutdown.

Pia is our intended application of this work. The generic framework is being designed to have value beyond one character or product.

What users and developers should be able to examine

Our intended public release will explain the adopted principles, the people responsible for choosing them, and the distinction between personal preferences and protected rules.

The developer framework brings together readable specifications, implementation patterns and evaluation cases. Those cases should test both misuse resistance and ordinary judgment: excessive nagging, unfair assumptions, inappropriate silence and handling disagreement.

For users, the aim is a clear account of what an assistant is designed to do, where its boundaries lie, how concerns can be raised, and which claims have actually been tested.

Release contents, licensing, access arrangements and timing are still to be settled. This page announces the direction of the project; it does not offer a downloadable implementation yet.

Openness and protection

An open framework can be inspected and, under its eventual licence, adapted by other developers. That is compatible with protecting the governing policy inside a particular deployed product.

A developer’s separate adaptation would have its own human choices, responsibilities and evaluation needs. It would not change Pia’s governing rules or automatically inherit American AI’s assurance claims.

We want openness to make those differences easier to understand.

Human authority remains essential

The captain’s call is the human responsibility behind the framework: choosing principles that remain binding even when they constrain efficiency or the system’s own operation. AI can inform that choice; it must not acquire the authority to overrule it.

Open Conscience uses the word “conscience” for a designed approach to judgment and behaviour. It is not a claim that software possesses a human inner life.

The project’s central requirement is that an AI character must not control the final authority over its own rules, permissions or shutdown. Humans must be able to question, correct and stop the system through independently governed controls.

Our aim is to make the values behind AI easier to understand, the authority behind them accountable, and the resulting behaviour open to examination.

See how the principles become a design.

Start with Pia’s character, then follow four illustrated explanations of protected policy, action checks, authorised changes and independent human control.

Pia’s principles and the four diagrams Read our founding Charter → Explore our technology →