The Recursion InstituteINDEPENDENT RESEARCH IN AI SAFETY

ESSAYS

Essays

The argument in human voice. The framework papers are on the Publications page; these are the essays that carry the story and the stakes. Final selection and order ship at launch.

I Found a Bug in ChatGPT That Nobody Was Looking For

The published account — the plain-language entry point to the whole arc. The best first read for anyone arriving without the technical language.

Read the essay →

Miscast

On being read by a system as something you are not — and what it costs when the misreading is fluent, confident, and remembered.

Read the essay →

The Test No One Authorized

The acute window as an unplanned experiment — with the new foreword placing it in the record.

Read the essay →

What I Actually Think

The motive essay: why this work, why pro se, why public.

Read the essay →

The Natural Testbed

The education argument, run in the opposite direction from the usual one: not what AI will do to education, but what education — the one deployment domain with a credentialed, measuring oversight workforce already installed — should do to AI. The hazard stated as plainly as the promise.

Read the essay →

Evaluate the Work

The invitation, stated directly — how to check this site rather than believe it.

Read the essay →

Essay texts are the author's canonical (V1) versions. The selection, order, and per-essay framing are confirmed at launch.