The Privacy Position

Privacy is not a feature. It is the precondition for an AI that serves people instead of harvesting them.

When you talk to a model, you are handing it the most personal artifact you produce: your live thinking. Search queries gave away your curiosity. Social feeds gave away your attention. Models give away your reasoning. If we let the AI era be built on the same exfiltration defaults the social era normalized, we have not solved anything. We have just changed the size of the bucket.

The first design constraint of a serious AI product is: what does this system know about its user, and who else knows it?

Where the privacy position is non-negotiable

User-owned memory. Sovereign / local-first options for sensitive work. Audit-friendly logs the user can read. No silent retention. No model training on private interactions without explicit, revocable consent. No surveillance as a business model.

These constraints are not anti-business. They are the only honest foundation a knowledge worker, a leader, a journalist, a lawyer, a clinician, an artist, or a child can build a real working relationship with a model on.

This is a draft seed of the full letter. The complete edition will land in the published series.