Useful enough to act. Governed enough to trust.
We write down what AI may read, draft, change and send, and we keep a named person accountable for what it does. On Claude, ChatGPT or Copilot.
Ethics is not another prompt.
Responsible AI does not live in a sentence typed into a chat box. It lives in the data the system can reach, the permissions it holds, the workflow around it, the person who reviews it, and the record it leaves behind.
Delegation does not erase responsibility.
You can delegate research.
You can delegate drafting.
You can let an assistant act inside your systems.
You cannot delegate the decision about how much authority it gets.
That decision stays with the firm, and it should be written down.
Before an AI recommends or does anything that matters, five questions.
We use them on every build.
Context and purpose
What is this actually for, and who does it affect?
Proceed only when the purpose is clear and legitimate.
Limits and authority
What may it see, decide or change, and who said so?
Nothing consequential outside the authority it was given.
Evidence and truth
What do we know, what is inferred, and how sure are we?
Verify material claims. Never fabricate a source.
Agency, dignity and fairness
Does it respect the person on the other end?
Redesign or escalate when it does not.
Risk and repair
If it is wrong, how will we know, and how do we fix it?
Choose actions that can be reversed. Review more as stakes rise.
Each answer comes out Green, Amber or Red.
Green proceeds.
Amber pauses for evidence or a person.
Red stops.
No single model proposes, approves and carries out the same consequential action.
Authority map · Hartwell Advisory · client follow-up assistant.
Illustrative.
Reasoning can be broad. Authority to act should be scoped.
Founding reviews, by conversation.
We are applying the framework with a small number of firms before we publish a fixed scope and price. A review looks at the AI your firm already uses: what it can reach, what it is allowed to do, who checks it, and what happens when it is wrong. If your firm uses AI with client information, we would like to hear how.
Start a conversationMapped to:NIST AI Risk Management Framework·ISO/IEC 42001·ISO/IEC 42005·OWASP GenAI Security Project guidance
Standards tell us how to govern risk. Ethical reasoning helps decide what responsible use of power looks like. We use both. We map our work to these standards; we are not certified to them, and neither is anything we build.
This is not legal or compliance advice. Your obligations are your own.
Where our framework comes from
The Clarity Ethical AI Framework began with Jewish ethical teaching, with a Messianic Jewish interpretive layer: human dignity, truth (emet), justice, not placing a stumbling block before someone who cannot see it, a duty to protect, authority as stewardship, and repair after failure (teshuvah). We apply these as principles for governing AI, not as religious rulings. The technical controls are kept separate from the values they serve, so a firm with a different tradition, or none, can bring its own commitments without weakening the controls.
Clarity Systems Lab does not issue religious rulings or certifications. The framework has not been reviewed or endorsed by any religious authority.
Three questions firms ask first
Does this make AI less useful?
No. Reasoning stays broad. What we limit is what it can do without a person, and only where the stakes call for it.
Is this a compliance review?
No. We write down what the AI may do and what is checked. Your compliance officer decides whether that is enough.
Do I need to share your faith to use it?
No. The controls are the same for every firm. The framework’s roots are stated so you know where it comes from.