Your AI model vendor's agreement does not cover your product

A while back I was talking to an AI app developer. He had built a number of apps on models he found on the web. Some came from reputable companies, some were just freeware. He wanted to sell a few of these apps to hospitals.

I told him that hospitals and other medical facilities work under strict compliance rules. Any software they use has to meet those rules, and that includes the AI models and agents inside it. I asked him what he had done to meet them, because without that the hospital would not buy from him. He said the service level agreements he had with the AI model providers should cover it.

That is one of the most common misunderstandings I see in startups and small companies. They rely on the model vendor's service level agreement. The freeware models do not even have one, so for those that answer was never there. For the others, the agreement covers the vendor's service. It does not cover how your product uses the model, what your app sends to it, or what your AI says to a patient. Either way, a service level agreement is not enough to meet the obligations a regulated industry puts on you.

When a hospital, bank or insurer buys an AI product, their security team sends the vendor a questionnaire. These days it has a section on the AI itself: how you test it, who watches it after launch, what customer data goes to an outside model provider, and what happens when it gives a wrong answer. For a small company, that is often where the deal sits and waits.

We put together a free AI Review Readiness Check for that moment. It is ten plain questions and takes about three minutes, with no sign-up. At the end you get a score, the questions a reviewer is likely to ask you, and what a good answer looks like. The picture at the top is a sample result. Your answers stay in your browser unless you choose to send them to us.

Getting approved is only the first step. Models and providers change quietly, and next month your customer will want to know the AI still behaves the way it did when they signed. That is the part we built Aggi ARIA (Aggi Responsible Intelligence Assessor) to handle. It tests how your AI behaves, as often as daily, and keeps dated reports you can hand to a customer. The Aggi ARIA AI Health Check (ARIA Shape B) costs $30 a month. We work with small teams directly, and the people answering your questions are the ones doing the testing.

I think the check is most useful for founders and CTOs selling AI into healthcare, banking, insurance, HR or government. Advisors and auditors who walk companies through these reviews may find it handy too.

I have a horse in this race; this is what my company does. The check is useful whether or not you ever talk to us.