ioLiteLabs, Inc.

Story

Field notes from
the evaluation layer.

Essays and arguments on how AI behaves when the conversation turns dangerous — and why no one is measuring it yet.

Policy·October 4, 2026·7 min read

The Pledge Is Not the Proof: Why AI Safety Now Needs Independent Audits

Six frontier labs pledged to hire independent auditors, and there is still no shared standard for what those auditors should measure. A pledge is a starting line, not proof. Here is what the week means, and what real independent auditing should look like.

Read essay
Policy·September 27, 2026·9 min read

Share the evidence, not the playbook: what the AI slowdown lawsuit teaches the safety community

Researchers can talk safety across company lines; they just can't be the channel their companies coordinate through. What the AI slowdown lawsuit and the law around it teach the safety community, and why the answer is a public safety commons.

Read essay
Commentary·September 23, 2026·5 min read

Fifty-seven percent: my read of OpenAI's MentalHealthBench

OpenAI's new benchmark is good work, and its best model still scored 57.3% on mental health conversations. What the number says, what it cannot say, and why independent checking has to be continuous.

Read essay
Open Call·June 10, 2026·2 min read

A Genuine Call for Partners

An open letter to everyone who touches AI — believers, skeptics, builders, parents, and yes, even the warmongers. A safer future needs all of you.

Read essay
Founder's Note·June 7, 2026·3 min read

Hello World, iolite here!

The first thing you ever print when you learn to code is 'Hello World.' So here's ours — and the very personal reason this lab exists at all.

Read essay