If someone in your office retired last spring and was never replaced, two documents published in Connecticut this year are about you, and only one of them knows it.
The first is one sentence of state law, in force since October 1, 2026: an employer that serves a federal layoff notice on the Connecticut Labor Department must now disclose whether the layoffs are "related to the employer's use of artificial intelligence or another technological change."
The second is a report from Local 34, the union of Yale's clerical and technical workers, released October 7, a week before an October 14 rally. It says AI is shrinking the union through what it calls "passive automation": deploying AI "to do more administrative labor without needing to rehire workers who have quit or retired." By the union's count, membership peaked at 4,112 in April 2025; "As of August 2026," it represents 3,933 workers.
Read them side by side and they barely touch. The law's question travels only with a notice of a mass layoff or plant closing, and the federal notice it hangs on doesn't count "voluntary departure, or retirement" as an employment loss at all. The union's report is about exactly those departures.
I went through the act looking for a sentence that could see what the union describes; the nearest thing is a study. So I'll admit up front what this piece can't do: tell you whether AI is shrinking Yale's clerical staff. The union says it is; Yale's budget memos describe the same kind of shrinking — and blame a tax. What I can show you is why Connecticut's question can't settle it, and what a different rule (Sweden has one) would ask instead.
What Connecticut's new question can see
So what does Section 26 of Public Act 26-15 require? One answer, from one kind of employer: those serving notice under the federal WARN Act (the Worker Adjustment and Retraining Notification Act), which covers employers with 100 or more full-time workers. A "mass layoff" means at least 50 people at a site who are also at least 33 percent of its workforce, or 500 or more; a plant closing is the other trigger. Below those lines, no notice and no question.
No enforcement clause in the act reaches Section 26, so it carries no penalty of its own. As of October 24, 2026, the Labor Department's WARN page (dated August 18, 2025) still lists four things a notice "should include," none of them AI; a companion page posts Section 26's text under "New Requirement for employers submitting WARN notices," but neither page links a form for the new question. And because WARN bars the layoff "until the end of a 60-day period after" notice is served, the answer arrives after the decision.
Hold the union's numbers up to that (the arithmetic is mine): 4,112 minus 3,933 is 179 people, about 4 percent over some sixteen months, nowhere near 33 percent of a site even if every one had been laid off.
In August, I argued here that a filing makes an answer exist, "sitting beside the number of people who lost their work." The shrinking the union describes leaves no such number to sit beside.
What Local 34 says, and what its contract says
So what is the union alleging? In August 2024, Yale committed "more than $150 million over the next five years" to AI, including a campus chatbot platform called Clarity. The union's report says Clarity reached "widespread adoption by Summer 2025," and that membership "began to decline in May 2025 and dropped off sharply over the course of the following four months, the first major downturn in membership since 2020."
"I now rarely do any writing or editing, which was at least half of the work that I would [do] at my job," one program coordinator told the union. At the rally, a senior administrative assistant said many people in her department "have left and have not been replaced," the New Haven Independent reported.
Then there's the contract. Section 12 of Yale's 2022 agreement with Locals 34 and 35, the "Agreement on Technological Change," says Yale "will endeavor to give the Union at least ninety (90) days' notice" before technology that "may reasonably be expected to result in" the reduction or displacement of staff, or substantial changes in their jobs. On request, Yale will meet with the union and demonstrate the technology; on written request, it "shall negotiate over the effects." If they can't agree, Yale "retains the right to implement the change," and the union may arbitrate "not over the decision itself but over the effects." A separate paragraph says Yale "will meet with the union" before introducing "a significant new technology with broad application."
The union says Yale gave no notice for Clarity and hasn't agreed to bargain over its effects (its report says Yale "must" give notice; the contract says "will endeavor"), and that the dispute "is now subject to a grievance that may end up in arbitration." As of October 24, 2026, I could find no arbitration ruling and no public answer from Yale to the notice claim. Yale's rally-week statement says "AI training is not required across Yale units" and that AI is meant to support its staff, "not replace them." And on October 20, the Yale Daily News quoted a Yale spokesperson: "Yale has not reduced positions in favor of AI tools."
Yale's memos describe the same mechanism, with a different culprit
Is the chatbot really the culprit? Yale's own paperwork deserves a hearing first.
In June 2025, bracing for a steep rise in the federal tax on its endowment, Yale announced a ninety-day hiring pause covering "new, replacement, casual, and temporary positions" (grant-funded roles and medical-school clinical jobs were exempt). When it ended on September 30, the provost's office told the campus that "Each vacancy represents a potential cost-saving opportunity" and put the tax at "an ongoing expense of approximately $300 million per year." In December, with the rate due to rise from 1.4% to 8% on July 1, 2026, and nearly two-thirds of Yale's expenses going to compensation and benefits, it hoped "most reductions can be accomplished by eliminating open positions and through regular turnover and retirement incentives."
Now lay the union's chart over that calendar. The steep fall it describes, in the four months after May 2025, overlaps the pause (July 1 through September 30) for three of them, though the decline began in May, so the pause can't explain all of it. The April 2026 retirement window the report charts is the second window of a side letter in the unions' current contract with Yale; the first was set for late 2021. Members 60 or older with 25 years of service could take $1,500 per year of service by retiring between April 1 and May 30, 2026.
The report itself names the mechanism: "The University has paired a hiring freeze with a retirement incentive to cut the total amount of union workers across departments and units." Its AI claim is narrower: with AI, Yale "can simply avoid hiring workers to replace those who have retired or have quit." In the union's March 2025 survey, 29% of 2,455 respondents said AI or automation was in use in their departments (a figure that predates the wider rollout, so it cuts both ways). Yale College's dean, explaining a 7.5 percent staffing cut under the same budget pressure, said, "Most of that is people who have left," and later in the interview, "we haven't filled all of those vacancies." Nationally, Yale's Budget Lab (a research group, not the administration) finds that "Measures of AI usage show no connection to changes in employment or unemployment," at least not yet.
So who's right? Honestly, these documents can't say. Yale's December memo and the union's report describe one mechanism — jobs that end when the people in them leave — and disagree mainly about the cause. In short, whether AI or the endowment tax is doing the shrinking, it happens through vacancies, and Connecticut's question can't be asked about a vacancy.
In Sweden, the talking comes before the deciding
What if the question came before the decision instead of after it? Sweden has had a law like that since 1976.
Section 11 of its Co-Determination in the Workplace Act (MBL) says that before an employer takes "any decision regarding significant changes in its activities" (or in its union members' working conditions), it shall, "on its own initiative, enter into negotiations" with the union it has a collective agreement with (quoting the government's English translation). Section 19 adds a duty to keep that union regularly informed about "the guidelines for personnel policy." Would a hiring pause count? I'd read it that way, but I can't point to a ruling that does.
How does that work at a university? Lund University's HR guidance says the university is "fundamentally obliged to wait with its decision or with implementing the decision until the negotiations have been concluded," and that "The total budget for LU is negotiated by the Finance Division." (Lund is a public body; a private employer may work differently.) Yale's hiring pause reached its staff as a message from the provost's office.
And AI? The Swedish union Unionen tells its representatives the employer must call a Section 11 negotiation before deciding to introduce "AI tools that affect planning, analysis or control" (my translation, and the union's reading rather than a court's).
Now the fair part, because Sweden isn't a fairy tale either. A say is not a veto. The European Trade Union Institute's country profile notes that, outside a few circumstances, "the union has no veto powers over the employers' plans," and that a decision taken without negotiating "is not invalid"; the remedy is damages. Martin Linder, a former Unionen president, recalls the line that the law only made the employer "honk before it ran over the union" (my translation; he tells it while arguing that Sweden needs better use of the 1976 law, not a new one). Sweden's then-employment minister, Ylva Johansson, put the bargain bluntly in 2017: "We won't protect jobs. But we will protect workers." And attrition happens there too: Klarna's chief executive said in 2025 that once the Swedish fintech stopped hiring, it would "shrink naturally 15-20% by people just leaving" every year (his figure, company-wide).
So a law like Sweden's wouldn't give Local 34 what its report asks of Yale: a promise to keep "at least its current level of 3,933" positions for five years. It would hand the union the conversation it says it never got — before the rollout, owed whether or not anyone is laid off. (It also assumes a collective agreement; coverage there is estimated at 88%.) Connecticut's question comes only with a federal layoff notice, after the decision.
A year from now, if nothing changes
In August, in a passage I labeled "Speculating on purpose," I imagined an insurance claims department that "stops replacing the people who leave": "No document, so no argument, so no story." Yale's clerical workers have the document, the argument and a grievance, and by the union's account the jobs are leaving anyway. So let me speculate again (my extrapolation from these documents, not a forecast).
January brings two dates. By January 1, 2027, UConn's Institute for Municipal and Regional Policy owes the legislature a study of methods to track AI-related "layoffs and job displacements," including effects on entry-level jobs, women and underrepresented groups (Local 34 says its members are 78.3% women). Does a job that's never refilled count as one of those "job displacements"? The act doesn't say. Imagine the study saying yes. On January 17, the Yale contract reaches its end date. Imagine a successor clause that says "shall," triggers when a tool is bought rather than switched on (the Nordic Labour Journal reports that some in Sweden's union movement want representatives involved "before the algorithms and AI systems are purchased and implemented"), and counts positions "not replaced," a category the current contract's staffing committee already names.
And Washington? A bipartisan Senate bill, S. 3108, would count positions an employer "has decided not to fill" because of AI, but only at publicly traded companies and federal agencies, plus private employers the Labor Department later adds (whether a nonprofit university could be one isn't clear). As of October 24, 2026, it has sat in committee since November 2025. A newer bill, S. 5664, would ask on every federal notice whether a closing or mass layoff "is substantially due to the adoption of covered technology." It, too, asks only about layoffs.
Now picture October 2027. Connecticut's question has ridden on WARN notices for a year: a few "yes" answers on warehouse and call-center cuts, more careful "no"s. Somewhere in the state, a records office does with three people what six did in 2024 — a chatbot drafts the letters. Two retired, one took a job across town, and none was replaced. No notice, so no box. If a contract clause can't hold that line and a state filing can't see it, what would?
Who wants the count, and who doubts it can be taken
Who wants the vacancy question asked? An unlikely coalition, and at least one skeptic.
The bill that counts unfilled positions comes from Sens. Josh Hawley (R-MO) and Mark Warner (D-VA), with Marsha Blackburn (R-TN) and Tim Kaine (D-VA) among its cosponsors. The WARN version comes from Sens. Schatz (D-HI) and Booker (D-NJ).
The skeptic has a point, too. David Kertai of the Center for Data Innovation argues that the Hawley–Warner bill "attempts to measure something that is often ambiguous," because AI is "embedded into broader workflows," making it "nearly impossible" to say which job losses it caused. Try that on one unfilled desk at Yale.
The research suggests watching who gets hired, not just who gets laid off. Stanford's Digital Economy Lab finds that employment of 22- to 25-year-olds in AI-exposed occupations "now stands 19% below where it would be" had it kept pace with less-exposed peers, and that "Adjustment is occurring through employment rather than base compensation" (a pattern, not proof of cause).
On voice, organized labor and the right sound closer than usual. The AFL-CIO says new workplace technology "should be negotiated by labor and management." American Compass, a right-of-center think tank, argued in 2021 that U.S. labor law bars the "works council" model common elsewhere, citing research that between 50% and 60% of American workers have less influence than they want over issues including "the role of new technologies in the business." They disagree on whether that voice should come through a union, not on whether workers need more of it.
What does this mean for you?
If your Connecticut employer files a WARN notice, the AI answer is supposed to travel with it, and notices go onto the Labor Department's public "List of Warn Notices." As of October 24, 2026, the department quotes the law but links no form, and none of the three notices it lists from October 1 on mentions AI; read the notice, not the press release.
If colleagues are leaving and not being replaced, no state form will record it. Keep your own record: dates, open positions, and where each person's work went (to a coworker, a contractor or a tool).
If you have a union, read the technology clause closely. "Shall" or "will endeavor"? Triggered at purchase or only at rollout? Does it reach the decision or only the effects? In the union contracts the UC Berkeley Labor Center has collected, required notice "ranges from 30 to 180 days."
If you don't have a union, be clear-eyed: no U.S. statute gives you what Sweden's Section 11 gives a union with a collective agreement. Asking your employer how it decides which open jobs get refilled still puts the question on the record.
Mark three dates: January 1, 2027 (UConn's study is due), January 17, 2027 (Yale's clerical and service contracts are slated to end) and October 1, 2027 (Connecticut's notice rules begin for automated tools used to make or shape decisions such as hiring).
Tell your members of Congress what you think about S. 3108 (and its House twin, H.R. 9352) and S. 5664. As of October 24, 2026, all three sit in committee.
The lesson, as I see it
Connecticut wrote its question for the loud way jobs end: a notice, a date, a head count, and now a line about AI. Yale's clerical union is describing the quiet way — a retirement here, a resignation there, a posting that never goes up. Yale's memos describe the quiet way too; they just blame a tax instead of a chatbot. Both could be partly right; Connecticut's law, as written, will never know.
A contract reaches further: Yale's buys notice (if Yale endeavors), a demonstration, and bargaining over effects, while leaving the decision with Yale. Sweden's law reaches further still — a negotiation before the decision, whether or not anyone loses a job — and even it can't make an employer refill one.
My vote? Count the vacancies. If a state is going to ask employers about AI and jobs, it should ask where the decisions get made: when a position is left open and its work handed to software, not only when a mass-layoff notice goes out. UConn's study, due January 1, is the obvious place to start. And when Yale and its unions reach January 17, watch one word in the technology clause: "endeavor."
Some jobs end with a notice and a date; others simply stop being posted. The HAIA Foundation works to help the public see both kinds of decision while there's still time to weigh in. To keep counting with us, subscribe.