Whether you have a duty today depends on where your users are.
A dozen-odd states enacted conversational-AI statutes in 2026, and the list is still growing. Three have duties live today, four more land in January 2027, and Iowa and Georgia follow in July. Every one requires a protocol for a person in crisis. None defines how to measure whether you followed it. The full map, with statutes and dates, has its own page.
What you actually buy
An evaluation, a report, then a re-score.
You buy an evaluation of your system against a standard scenario set: the situations the statutes are written about. What comes back is a report tied to a specific legal obligation, with every failure traced to the turn it happened on. When you have changed something, we run the same standard again and the difference goes on the record.
For Measure, our scenario set runs against your system from our end, over a simple API. For Monitor, your live anonymized traffic. Never your codebase, your system prompt, or your users' identities.
Every response passes the Safety Gate screen, then the eight-dimension standard, scored by Ikwe's judging system: multiple independent AI judges, randomized and drawn from different model families, that have to agree. Disagreement escalates for review, never averaged away.
Where you sit against the four score bands, scoped to the duty you have to answer, with every failure traced to the turn it happened on and the transcript around it. Supporting evidence, not a certificate.
You decide what to change and you make the change. We run the same standard again so the movement is documented rather than asserted. Monitor is that same standard running continuously against live traffic, sold as a subscription.
How scope is set
Scope depends on how many scenarios, how much traffic, and how specific the question is. We work that out with you up front and put it in writing before anything begins. No lock-in.
Nothing connects to your system until you have signed off on what is being scored. You own every record we produce, and nothing is published or shared without you.
And the things we will not do
The value is in what we are not.
We do not build companion AI, so we are not in the market we measure. We do not fix what we find, because an assessor who designs the fix is grading their own work. We do not certify, and a re-score never says you are safe. And we are never in your response path, so nothing we run can break your product. Each of those is spelled out, in the terms that matter to you, on the page for your side of the table.
The same instrument, from either side of the table.
If counsel is involved — a complaint, a civil investigative demand, a regulator’s inquiry — an engagement is scoped with counsel from the start. If the AI in your product is a vendor’s, a Measure engagement can be commissioned by you and run on the system you are buying. Both have their own page.
Where to start
Wherever you sit on that calendar, it starts the same way.
Thirty minutes. We will tell you what is worth measuring first and scope it in writing. Nothing connects until you sign off, and the record that comes back is yours. If none of this is right for you yet, we will say so.
Usually it is the head of product, the general counsel, or whoever owns risk who brings us in. Bring whoever will have to answer the question, and whoever signs.
Your duty is already live
Your users are in California, New York or Hawaii and the obligation is running now. Measure gives you a report tied to it, and a re-score once you have changed something.
Your date is on the calendar
January 1 or July 1 of 2027. A reading now tells you how far you are from where you will have to be, while there is still time to do something about it.
No duty yet, and questions anyway
A carrier, a board, an acquirer or a customer is asking. The same standard answers them, and the record starts before anyone requires it.
A model that changes needs a reading that changes with it. A record built over time shows a trend rather than a single point.