
What Should an AI Be Allowed to Remember?
A product-builder’s analysis of AI memory, inferred identity, deletion, and a practical contract for governed retention.
Reading path
Understand how AI memory can support continuity while also turning old facts and inferred preferences into pressure on the person someone is allowed to become.
Why this order
The path first separates storage from humane memory, then follows the consequences when a system's accumulated model of a user becomes more coherent than the user's relationships can be.

A product-builder’s analysis of AI memory, inferred identity, deletion, and a practical contract for governed retention.
Begin with the mechanics and ethics of retention so the difference between stored facts, retrieval, inference, and forgetting is clear.

A practical argument for making AI personalization contestable through provenance, scope, expiry, correction, and influence audits.
Continue to the social consequence: a detailed machine profile can influence identity even when every remembered detail is individually accurate.