In July a hospital system in the Bronx eliminated twelve utilization review nursing positions after bringing in software from a vendor called Datavant.1 The nurses' union filed a class action grievance and issued a statement saying the hospital was replacing human workers with artificial intelligence. Hospital leadership called that characterization misleading. Challenger, Gray and Christmas, the outplacement firm that has counted announced American job cuts every month for decades and earns its living placing the people who get cut, went looking for a category to put those twelve positions in, could not get clarity from the hospital about what the software actually does, and filed them under a heading it keeps for exactly this situation: technological update, possibly AI.1 Twelve jobs, one event, and nobody in a position to know would say plainly whether a machine was the reason.

The Question Connecticut Starts Asking on October 1

That kind of vagueness has been free. In Connecticut it stops being free at the start of next month, when Public Act 26-15 takes effect.2 Buried in a long statute covering hiring tools, chatbots and frontier model developers is a short provision aimed at layoffs. Any employer whose reduction is large enough to trigger a federal WARN notice must also file a written disclosure with the state Department of Labor saying whether the layoffs are related to the employer's use of artificial intelligence or another technological change.2 The employment attorneys at Epstein Becker Green, reading the act closely, make a point that is easy to miss: the obligation is not confined to whether a model helped choose who got cut. It reaches whether the technology contributed to the reduction at all.3

Which is a considerably harder question, and one most companies have never had to answer with a name attached to it.

The Reason Gets Chosen After the Number

Look at the volume this is landing on. Through July of this year Challenger counted 112,713 announced job cuts in which artificial intelligence was given as the reason, roughly a quarter of every cut announced in 2026, and AI has led the list of stated reasons for five months running.1 Andy Challenger, the firm's chief revenue officer, laid out the calculation with no decoration at all. Naming AI in a layoff announcement plays well with investors while pushing current and prospective employees away, and that tension is why the messaging moved from hedging to citing it aggressively.1

Notice what he is actually describing. The reason a company gives for a layoff is being selected according to its effect on an audience. That is a communications decision, made after the headcount decision, often by people who were nowhere near the room where the headcount got set. I have been in enough organizations where the number arrived first and the explanation was assembled afterward to know the order those two things usually happen in, and the explanation is rarely the one that comes first.

An organization that cannot name which work moved to the machine has not made a decision about AI. It made a decision about headcount and then borrowed a reason that tested well.

The Filing and the Business Case Want the Same Evidence

Here a compliance chore turns into something an executive should care about. To answer the Connecticut question honestly, an employer has to be able to point at the work. Which tasks moved to the tool. Who checked that they actually moved rather than assuming it from a demo. What the tool now does that a person used to do, and what was left sitting there afterward with nobody assigned to it. None of that is legal work. It is the operating analysis that should have run before the cut, not after the notice.

It also happens to be the same analysis that would let a company claim a return. Ask nearly six thousand senior executives across the United States, Britain, Germany and Australia what AI has done to their own firms and more than nine in ten report no effect on employment over the past three years, with 89 percent reporting none on labor productivity. Sixty-nine percent of those firms are actively using it.4 I went at that gap between the forecast and the lived experience a few weeks back in the piece on executives predicting a boom they have not yet seen, and I am not going to relitigate it here. Today's point is narrower. A company that cannot evidence an employment effect for a state labor department cannot evidence one for its own board either. It is a single missing piece of work, and it is about to be visible from two directions at once.

There is an obvious way to dodge this, and Challenger expects companies to take it. As the rules take shape, announcements will get more careful and the real effect of AI on jobs will get harder to see from outside.1 That response is rational and very small. It protects the filing while leaving the actual problem sitting where it was, which is that the business still does not know what its own technology did to its own work.

The duller response is the useful one. Before the next reduction, write down the causal claim you would put your name on. If you can write it, the disclosure is a formality and your investment case has something under it. If you cannot write it, you have found out something more valuable than a compliance answer, and you have found it out before a state agency, a union, or a reporter goes looking for the same thing.

So, for anyone holding a workforce plan this quarter: if a labor department asked you today whether technology contributed to it, what would you write down, and who in your building could stand behind it?