The law catches up, slowly

No federal layoff notice asks why, and the one state law that does asks so broadly that a yes can say nothing about AI.

New York asks on a form the public cannot see, and one notice out of 214 this year names AI.

The stronger rules are still waiting: a New York bill one signature from law, and a Senate bill untouched in committee for almost a year.

"Artificial intelligence is already replacing American workers." Senator Josh Hawley said that in November 2025. If it is true, it should show in the government's own record of layoffs.

So we went looking for the question: the line on an official form where an employer must say why the jobs went, and whether AI played a part.

Here is what we found. The federal law never asks. One state asks, so broadly the answer means little. Another asks on a form almost nobody can see. A stronger bill waits for a signature, and Congress has done nothing. Europe went first.

When a large American employer cuts jobs, a federal law called the WARN Act makes it warn the state first, in a notice filed weeks ahead: how many people will go, and from which building. Those notices are the only place any government counts a layoff by site and by date. Anyone can read them.

They are also where a layoff becomes real. Oracle never said how many people it cut this autumn. The state lists did: 800 people, all leaving on 13 November. That was 63 in Pleasanton, 279 in San Mateo and 99 in Santa Clara, in California, and 359 in Seattle.

California's list alone carries 397 notices, covering about 22,600 workers, in the four months to late October. Of those, 180 notices, about 13,100 workers, take effect on or after 29 October. Those are people still working out their notice today.

What none of those notices said, until this year, is why. The federal law does not ask.

Since 1 October 2026, Connecticut has required any employer filing a layoff notice to tell the state whether the layoffs relate to its use of AI "or another technological change." It is the first such rule written into a state law.

Two things blunt it. The question is wide: "another technological change" covers almost anything, so an employer can answer yes truthfully and tell the state nothing about AI. And the rule is hidden in plain sight, one section of a long law called "An Act Concerning Online Safety."

New York got there first, in a different way. It never changed its law. Its labour department added a question about automation and AI to a form inside a portal that needs a login. The wording has never been published. The state's public dashboard shows no reason for any layoff, though the data export behind it carries one.

So the two counts in circulation disagree. A law firm reported in May that none of more than 160 notices had answered yes. The state's own export, which we read, shows one notice this year that names AI. Nespresso gave "Relocation of Business, Artificial Intelligence" as its reasons, for 46 workers in Manhattan. Out of 214 notices, one.

A disclosure nobody is required to make clearly, which nobody indexes and nobody counts, shows how far the record has to go.

The stronger rule may come from New York's legislature. Bill A9581-B has passed the Senate and the Assembly. It is waiting to be delivered to the governor and signed, as the New York Senate's page read on 30 October 2026.

If it is signed, every New York business with 50 or more employees would report each year how many jobs AI cost. That would include the positions it chose not to fill, which no layoff notice will ever capture. The penalty for not reporting would be $500 a day.

It would be the strictest rule in the country, and almost nobody has written about it.

In Washington, a bill from both parties would have the Bureau of Labor Statistics, the government's official counter of jobs, publish AI-related layoffs every quarter. Hawley, a Republican, introduced it with Senator Mark Warner, a Democrat, in November 2025. Warner's line was "Good policy starts with good data."

The bill, S.3108, has had one action in its life: it was referred to committee, the first step. Since then the cosponsors kept coming, Tim Kaine in January and Marsha Blackburn in March. Two Democrats, two Republicans, and the committee has not touched it once. Even if it passed, its text has no penalty, no threshold, and no definition of its central test, "substantially due to."

Jim Banks, who chairs the Senate subcommittee that would hear it, said in July the predicted mass displacement has not begun: "we haven't seen that begin to happen." In the House, Steven Horsford introduced his own version in June. "Behind closed doors, with no disclosure and no accountability, entire livelihoods are being erased," he said.

On 23 October 2026, Senator Ron Wyden announced a bill to widen unemployment insurance for workers displaced by AI, and another to tax data centres, as Roll Call reported. Neither has a bill number yet. Other bills would tax AI where it cuts jobs. None has passed.

Europe did not wait for the layoffs. The EU's AI Act requires employers to tell workers and their representatives before a high-risk AI system is used on them at work. High-risk here means systems that hire, promote, assign tasks, monitor or fire. It is the only binding rule of its kind in the world. A delay of those duties to December 2027 has been reported, but so far it rests on a law firm's alert, not the official text.

In Germany, a works council, the body that speaks for a company's employees, must be told of plans to bring in AI, and can call in an expert. Spain gives works councils the right to know how the algorithms that affect jobs work. South Korea's new AI law covers decisions about employment, but asks for no reporting of jobs lost.

The Securities and Exchange Commission, which polices what listed companies tell investors, writes to a company when it wants a claim explained. Since January 2025 it has sent four such letters to the 23 companies we followed, all to Oracle and Intel. Two only closed the review. By our reading, none asked about layoffs, headcount, or what the companies say AI does.

So, for now, a company can describe AI and its cuts however it chooses. What they choose to say is in our piece on what the companies say.

Three things, none of them hard. A question about AI on every state's layoff notice, worded the same way and published. An official count of the answers. And one federal question, on the WARN Act itself, so a layoff in Seattle and a layoff in Hartford are asked the same thing.

What we hold today is enough to show the gap: Connecticut's law, New York's form and its own export, and the layoff notice lists of the states in our record. Every count in this piece comes from those, and each one is only as wide as the record behind it.

There are two ways to read the silence, and the evidence supports both.

The first is that the record hides the answer. Employers are not made to say, so they don't, and one notice in 214 tells us more about the question than about the layoffs. That is Hawley's reading, and Horsford's: the jobs are going, and nobody has to account for them.

The second is that there is not much to find yet. The senator who would hear the federal bill says mass displacement has not begun, and one in 214 may simply be the true rate.

Either way, about 13,100 workers on California's list are working out their notice right now, and none of their notices says why. And the question is not going away: candidates in at least 45 states now mention AI, the Washington Post reported in August.

Our ask is simple. When you see a count of jobs lost to AI, ask who asked the employer, and what exactly they asked.