Colorado just blinked on AI and we’re going to pay for it

In 2008, the American financial system collapsed because a handful of institutions were packaging garbage into products nobody could see inside, selling them to everyone, and pocketing the returns while it lasted. Nobody was required to explain what was in the box. Nobody with a financial interest in the system wanted to look.

It didn’t fall apart. It imploded. Because every fund, every bank, every pension holding those products was exposed to the same underlying rot they didn’t know they shared.

We are doing it again. The box is just different now.

Today the decisions that shape your financial life — whether you get a loan, whether your insurance claim gets approved, whether you’re considered for a job — are being made by AI models that the institutions using them are not required to explain, and in many cases cannot explain, because they outsourced the thinking to vendors they’ve never audited.

These vendors — which most Coloradans have never heard of — are running models inside the credit unions and lenders and insurers you interact with every day, deciding your outcomes at scale with no public record of how or why. The same vendors. The same architectures. The same correlated risk, invisible, moving through the entire system simultaneously.

When those models are wrong — and they will be wrong — they will be wrong everywhere at once. And unlike 2008, there won’t even be a paper trail of what was inside.

I built AI-powered businesses generating millions in revenue, and reached a point where I couldn’t explain why my own models were making the decisions they were making. Naturally, I went looking for an independent transparency framework, found nothing, and built one. AI Clear rates how transparently companies disclose the AI systems making consequential decisions about people’s lives. The methodology is public. The ratings are open to anyone.

The pattern holds across every vertical we rate, and it runs the wrong way: the more consequential a sector’s decisions are for the people on the receiving end, the less it discloses about the AI making them. Insurers deciding whether to pay a claim. Lenders deciding whether to approve a loan. Health plans deciding what care gets authorized. Employers deciding who gets screened out. Across all seven regulated domains, not a single company scored higher than a B. The data is open at aiclear.org. Anyone can check where their own bank, insurer, or employer stands.

Colorado was the first state to try to force this open. Senate Bill 205, passed two years ago, pushed companies toward real explanation: the principal reasons behind a consequential decision, the reasoning a person could use to challenge an outcome that upended their life, rather than the bare fact that AI was in the loop. The principle that a system making decisions about you should be able to explain itself is what the field calls explainable AI.

In May, Senate Bill 189 repealed and replaced it. Companies now have to tell you AI was involved, describe the role it played in an adverse decision, and give you an appeals process. They do not have to tell you why the AI decided what it decided — the actual model logic a person could use to contest an outcome in court. A plain-language description of AI’s role is not the same thing as an explanation of its reasoning. “Everybody lost and everybody won,” said the bill’s sponsor.

The reason this happened is the same reason 2008 happened. AI is making institutions extraordinary amounts of money. A model that processes 10,000 loan applications in the time a human processes ten generates returns that make hard questions feel like bad business. Nobody inside these institutions wants to open the box. The economics don’t reward it, and now the law doesn’t require it.

That’s exactly where we were in 2006. The ratings said buy. The returns said buy. Nobody wanted to be the one asking what was inside. The box was never opened.

The data is public at aiclear.org and the rulemaking process under SB189 has until January 2027 to define what adequate disclosure actually requires. That window is still open.