Let me guess.

Your organization has an AI ethics policy. It mentions fairness, transparency, accountability, and probably privacy. It was drafted by legal, reviewed by a cross-functional committee, and published somewhere on the intranet.

Nobody follows it.

Not because people are unethical. Because the policy is unusable. It tells people what to VALUE without telling them what to DO.

The Ethics-Action Gap

Most AI ethics policies read like corporate philosophy papers. "We are committed to fairness." "We value transparency." "We believe in responsible AI."

Great. Now tell me: when your data scientist is building a model for loan approvals and they discover the training data underrepresents a demographic group — what, specifically, should they do? Who do they call? What form do they fill out? What authority do they have to stop the project? What happens next?

If your ethics policy can't answer those questions, it's a values statement, not an operating framework. And values statements don't prevent harm. Operating frameworks do.

From Ethics to Architecture

This is what I mean when I say integrity is an architecture, not a value statement. You have to BUILD the systems that make ethical behavior the default path — not the heroic path.

That means:

1. Decision checkpoints. At specific points in the AI development lifecycle, someone must answer specific ethical questions. Not "is this ethical?" — that's too vague. "Does this model's training data represent all affected populations proportionally?" That's specific. That's answerable. That's actionable.

2. Escalation paths. When someone identifies an ethical concern, there must be a clear, SAFE way to raise it. Not a suggestion box. A defined process with a named recipient, a guaranteed response timeline, and protection from retaliation.

3. Red lines. Pre-defined conditions under which an AI deployment must stop. Not "we'll discuss it" — full stop. "If the bias audit shows disparate impact above X threshold, deployment is paused until remediation is complete." Automatic. Non-negotiable.

A Prompt for Building Yours

"Draft an AI ethics decision framework for a [industry] organization. For each stage of AI development (data collection, model training, testing, deployment, monitoring), define: one specific ethical question that must be answered, who answers it, what happens if the answer is concerning, and what documentation is required. Make it practical enough that an engineer or analyst could follow it without interpretation."

That output will be more useful than 90% of the AI ethics policies currently in existence.

Try This Monday

Find your organization's AI ethics policy. Read it. Then ask yourself: could a team member who just joined the company use this document to make a real ethical decision? If not, it needs to become an operating framework.

I wrote about building these systems in The Architecture of Integrity — how to turn abstract values into concrete, operational structures that make your organization actually work the way it says it does.

Ethics isn't what you believe. It's what you build.

— Vance Sterling