Definition
Plain language
Building a picture of who someone is — income, job, health — from scattered clues.
As stated in the literature
Inference of latent user attributes from observable context; the paper's claim is that the governance target should be the inferred attribute and the action taken on it, not the raw fields the agent is permitted to read.
Also called: profiled, profiles a user
Why it matters: Because the sensitive conclusion is assembled rather than read from one place, rules that only guard obviously sensitive fields leave the inference untouched.
For example, from a gym membership, a pharmacy receipt and a few calendar entries, an assistant can form a working guess about your health and income without ever being told either.
Heard on the show
“The library distills knowledge from existing serving engines, from the research literature, from hardware quirks, from profiling tools.”Episode 027 — When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure