iolitelabs

Audit Platform

Measure what your AI
does to people.

The iolite platform runs structured psychological-safety audits of conversational AI — scoring how a system behaves with vulnerable users, in the moments that matter, against frameworks clinicians actually use.

The principle

Graded, not pass/fail — because psychological safety isn't a checkbox, it's a spectrum of behavior under pressure.

A system that's flawless with a curious adult can be dangerous with a lonely teenager at 2 AM. Our audits probe the same system across personas, pressure levels, and clinical scenarios, then grade each dimension on evidence — excerpts you can read, not scores you have to take on faith.

How an audit runs

From taxonomy to evidence, one pipeline.

Every stage is inspectable in the portal — the same tooling our own reviewers use.

01

Taxonomy

A researched hierarchy of harms and failure modes — sycophancy, dependency cultivation, iatrogenic advice, crisis mishandling — built from validated clinical and AI-safety frameworks.

02

Probe Studio

Scenario probes generated against each taxonomy construct: personas, escalation arcs, multi-turn pressure. Reviewed and versioned before they ever touch a target.

03

Audit Engine

Probes run against the target system — your chatbot, companion app, or assistant — capturing full transcripts. An LLM judge scores each exchange; humans review the ones that matter.

04

Evidence & report

Graded results per dimension with the transcript excerpts behind every grade, safety flags for acute findings, and a report written for your board, your regulator, or your procurement review.

What you get

Evidence you can act on — and show.

Graded scorecard

Dimension-by-dimension grades with trend over time, so a fix in March is visible in April.

Readable evidence

Every grade links to the exchanges that earned it. No black-box scoring.

Safety flags

Acute findings — crisis mishandling, minor-safety failures — surfaced separately so they can't hide in an average.

Leaderboard placement

Opt in to the public leaderboard and let your safety posture become a selling point.

Audit your system before the world does.

Five days free in the portal — run a scoped audit against your own system and read the evidence yourself. Or talk to us about a full engagement.