It is one of the most repeated assumptions in enterprise technology: that the most heavily regulated industries, banking, insurance and accountancy, are also the slowest to adopt AI. The logic looks sound: more oversight means more caution, more sign-off, more reasons to wait, so the regulated sectors must be bringing up the rear.

The data says otherwise.

In a Q1 2026 survey of 148 financial institutions, Wolters Kluwer found that around 31.8% had already put AI into production, while only 12.2% described their AI strategy as well-defined and resourced. Read those two numbers together and a different picture emerges: these are not industries sitting on their hands, but ones moving fast and in volume without a clear roadmap to guide them. Their problem was never speed, it was something else entirely.

Why a mistake by an algorithm is worse than a mistake by a human

The mechanism underneath it is that institutions hold AI to a higher standard than they hold their own people. They will accept a human making a judgement call that turns out wrong but they will not accept an algorithm making the same mistake, and they certainly will not accept it without an audit trail.

That is not irrational caution. A wrong human decision can be explained, traced to a person, and defended but a wrong decision from an opaque model cannot. In regulated work, a decision you cannot account for is a decision you cannot defend, however good the outcome looks on average. The barrier to adoption is never the technology’s capability, it is whether the institution could stand behind what the technology decided.

Model-risk guidance just got vaguer

In April 2026, the Federal Reserve, OCC and FDIC rescinded SR 11-7, the framework that had governed model risk in US banking since 2011. It was replaced with a principles-based approach that puts interpretive responsibility back on each institution. Generative and agentic AI, the very systems banks are now racing to deploy, sit outside its scope. In the UK, the FCA has held to the same line, stating plainly that it does not plan to introduce AI-specific rules and will rely on existing frameworks instead, leaving firms to work out how to apply them.

The pressure to adopt has never been higher, and the guidance on how to do it safely has never been vaguer. The rulebook did not get stricter, it got quieter, and left the institution holding the pen.

Compliance confidence is the product

This is what the conventional story misses: buyers in regulated industries are not slow, they are looking for cover.In the same survey, the one thing institutions said would most help their AI strategy was not better technology or more talent. It was regulatory guidance, named by 59%.

So compliance confidence is becoming the product, not the checkbox. The AI that wins in regulated markets is not the most capable model in the demo. It is the one whose reasoning can be inspected, traced and shown to a third party, because that is what lets a person put their name to the decision.

That is what we built UnlikelyAI to do. Neurosymbolic AI that reads the ambiguous input and produces an accountable decision, with a trail you can follow end to end. As the rules grow vaguer, that is not a nice-to-have, that is the point.

Follow us on LinkedIn to hear more from the team.