Back to the journalAI & systems / Field notes

Privacy starts before the model

Beyond wearables: a conversation about the infrastructure behind personal AI.

A fine ivory filament knot is enclosed by translucent glass membranes.
Personal data, held within deliberate boundaries.

Personal AI begins with personal data. In this interview with George Georgiev, Hakan Kozaklı discusses the connection between wearable devices, data ownership, and AI.

Begin at the source

The interview discusses signing data at measurement, Vyvo’s Proof-of-Sensing protocol, and the Life CoPilot concept. It also distinguishes assistance from medical diagnosis.

Those are claims made in the published conversation. This illustrated summary does not independently validate the product or its security properties.

01 / Three separate boundariesConceptual system view
  1. OriginCaptureWhere did it come from?
  2. PermissionAuthorizeMay it be used here?
  3. PurposeMinimizeWhat is actually needed?
A general way to reason about a personal-data pipeline, not a specification of Vyvo’s architecture.

Make the boundary explicit

At the application layer, broad access can often be narrowed to a few task-specific fields. The example below describes a session that uses a daily activity summary rather than raw signals.

context_policy.json
json
{
  "purpose": "daily_activity_summary",
  "allowed_fields": ["steps_total", "activity_minutes"],
  "raw_signal_access": false,
  "retain_after_session": false,
  "revocation_check": "before_each_use"
}

This JSON is an illustrative application policy, not a Vyvo configuration or an enforced security control. A real system must implement and test each boundary it describes.

Read the full interview

The original questions and answers provide the detail behind the discussion of data ownership, infrastructure, and personal AI.

Interview / George GeorgievRead the original publication

Sources & further reading