The Union Gets the Final Say
School districts have begun giving labor unions veto power over whether a promising technology may ever be used to increase the productivity of their workforces.
On June 25, 2026 the Seattle Education Association proposed that its school district “shall not implement, utilize or promote AI Systems that eliminate, reduce, or replace work traditionally performed by bargaining unit members without prior written agreement of the Association,” and “any productivity gains from AI System implementation shall not be used as a basis for staffing reductions or increased task and workload expectations.”1
Four months earlier, in California, the United Educators of San Francisco (UESF) carried similar phrases into a four-day strike — down to the definition of the systems covered, which in both documents reaches any “computer system, software, algorithm … or automated process” performing “pattern recognition … language processing … or data analysis.” That describes most modern payroll and benefits software.
San Francisco Unified School District signed a contract with OpenAI for up to 12,000 ChatGPT accounts just a few weeks before the strike but then accepted UESF’s AI restrictions in the midst of it. So, it purchased the tool, then agreed not to use it to change how the work gets done. Sacramento City Unified’s teachers hold the same guarantee in a ratified contract of their own.2
Los Angeles Unified School District (LAUSD) wanted a committee to sort things out. For months, it advocated for a joint task force to study AI and make recommendations. In March 2026, the chair of a fact-finding panel recommended that the district concede the AI consent requirement — and so it did. On June 16, 2026, its board approved in a single sentence: the “Establishment of a task force to discuss Artificial Intelligence (AI)-related issues including AI not replacing Bargaining Unit positions or members without the express written agreement of union.” A body was created to study the question but then handed the answer before it could schedule its first meeting. 3
During my thirty years in municipal management, we were accustomed to giving the union notice before changing how work was done. We would negotiate how a change would affect people — schedules, workload, assignments. That is effects bargaining, required by law for decades in the states that bargain with their public employees. But a manager required to give notice, or meet and confer, still gets to decide. A district management that must obtain written permission does not decide at all.4
The AI-related obligations are decidedly lopsided. San Francisco’s article requires the district to give ninety days’ written notice, documentation of what the system does, and an analysis of its effect on bargaining unit positions. Nothing in it requires the union to answer, sets a standard for a refusal, or provides an appeal from one. Seattle’s is the same. A teacher’s worry that a district in fiscal trouble will reach for AI to cut headcount is not irrational. But effects bargaining is in place to address such circumstances. The school districts’ AI concessions amount to an abdication of management.
Everyone Else Gets to Find Out
Employers elsewhere are overwhelmingly not firing people over AI. Gartner attributes about one percent of studied workforce reductions to AI productivity gains. Boston Consulting Group projects that 50 to 55% of U.S. jobs will be reshaped within two to three years: the same role, with very different expectations about how the work gets done.
That is the workplace question that is alive everywhere else — which tasks move, which roles change, what a job description should look like in three years. Nobody knows how it will turn out. Organizations learn by experimenting, not by negotiating permission to experiment.
Private schools face no such restriction on using technology to improve their outcomes and make their teachers more productive. Many of them will discover what AI can do and adopt it where it helps. These government school AI clauses are the taxi medallion of 2013 — a legal restriction that protects incumbents, raises the price, and holds only until the alternative arrives. Medallion owners bore their own losses. Public districts socialize theirs. The students who pay first are those whose families cannot afford another school. The taxpayers pay next, including those who leave and go on paying for the school their children left.
The Tutor Their Parents Can’t Buy
The greatest cost here is not administrative productivity as such. It is what students lose. If AI turns out to be even a fraction of the tutor it promises to be, the children who gain most are the ones whose parents cannot afford one. Just as the internet brought a world-class library to children without proximity to one, AI brings the possibility of a versatile tutor — one that keeps the precocious child from being anchored to the class average and gives the struggling child the help his parents cannot afford.
A school district that puts friction in the way of finding out what a promising technology is worth in a classroom is working against its own educational purpose. That it would bargain away the right to find out is unconscionable.
So They Cut People Instead
The districts giving up their right to manage the workforce are the ones that can least afford it. LAUSD enrolled 408,083 students in 2024-25, down from 746,831 in 2002. It is spending about $2 billion more this year than it takes in, and it has hired an outside consulting firm to build models for closing and consolidating schools, with consolidations deferred to the following school year. Seattle Public Schools has already committed its $42 million rainy day fund, borrowed $27.5 million from its own capital program, and closed a roughly $100 million gap — and warns that if its fund balance goes negative it faces binding conditions from the state, “potentially losing local control over how we spend money.”
An organization can control costs in only two ways: by changing what it does, or by changing how it does it. The AI restrictions neutralize the second in organizations far more likely to expand what they do than to trim it.
What that costs is on display at LAUSD. On June 16, the board adopted the AI memorandum without discussion, as part of $1.48 billion in labor agreements. The item it did discuss that evening was the Fiscal Stabilization Plan immediately preceding it — $2.488 billion in cuts and 6,008 positions, which the district had to pass first, or its chief financial officer could not certify that it could pay for the agreements. Two weeks later the county issued its first Lack of Going Concern determination for the district.
The potential to change the way the work gets done to make the teachers more productive is now frozen in the name of preserving jobs just as the organization was forced into massive personnel cuts.
These agreements further invert the district’s relationship between productivity and pay. Generally, technology makes a worker more productive, the output is worth more, and the worker earns more. That is a trade, and it is what makes a pay raise sustainable. Here the increase came first — an average of 13.86% — and the ability to execute improvements that could have paid for it is contractually prohibited. Nor may the district put the money where it would do the most good: teachers are paid on a single salary schedule keyed to training and experience, and departing from it also requires the union’s agreement. The excellent teacher is paid on the same scale as the one who is not — under the same permission requirement now extended to AI.
The district does not decide.
There is a structural contribution to this dysfunction. A private firm that refuses to adapt loses customers to one that does, so the employer and workforce have a mutual interest in productivity. A government school system faces no such test. Its revenue arrives as per-pupil funding. The people who bear the cost of operational stagnation are students and taxpayers, and they are not at the table. The organization across the table influences who sits on the board and is the gatekeeper to the tools of adaptation.
Nobody Had to Disclose
How does something as egregious as bargaining away one of the two means a district has for controlling cost get approved in a public meeting?
Because the only question anyone is required to answer is what it costs, from a very narrow perspective on what constitutes cost.
From what I can tell, the State of California has the most detailed process for approving a school district labor agreement. Before the board may vote, the district must lay the major provisions of the labor agreement in front of the public, and the superintendent and the chief business official have to formally certify that the district has the means to pay for it. Wages and benefits are calculated to the dollar and projected across the contract term. Everything that is not about money gets one line. On Los Angeles’s June 16 filing, that line reads “Not applicable.”5
Nothing in the process requires anyone to evaluate, quantitatively or qualitatively, the opportunity cost of losing control of the ability to improve work processes and work methods. Nothing requires the district to disclose that a key right of management will now be unavailable to the next school administrator and the next district board.
In Washington state, where the same restrictive AI clause is pending now, the requirements sit in a different place. A Seattle board member votes on the labor agreement in a public meeting, and his finance office gives him a cost figure — as a matter of the board’s own policy rather than state law. The state aims its attention at the budget rather than the agreement: every district must maintain a four-year plan estimating what it will cost to sustain current programs and service levels, and the state ranks every district by financial health to head off insolvency and the binding conditions that follow. [5]
Today, nobody can quantify with precision what AI might save a school district. But a district can recommend that its board forbid finding out. And in neither state does anyone have to discuss this when approving the agreements.
Three questions belong in front of every governing board before it ratifies a labor agreement: (1) what operational authority are we giving up, (2) what is it costing us to give it up, and (3) what would it cost to buy it back?
Districts already estimate the cost of a one percent raise. But the ability to manage the workforce is just as important, arguably more so. The two district officials who certify that the district can pay for the agreement should have to certify these answers as well.6
History will not record that AI failed America’s public schools.
It will record that we negotiated away the right to find out.
Seattle’s clause is a union proposal in open bargaining as of August 14, 2026, not an agreement. AI is not among the issues the parties have settled, the contract expires August 31, and the union’s lead bargainer says members will vote on a tentative agreement or a strike authorization by the end of the month. The union appears to understand how far its own definition reaches: one subsection would bar the district from uploading personnel, grievance, investigation, health or leave files to any covered system, and another carves out an exception for the district’s payroll and absence-tracking systems exchanging data with each other. A definition need not be enforced at its edges to govern behavior. It only has to be broad enough that counsel cannot tell a superintendent with confidence that a given purchase falls outside it.
San Francisco has not published the executed language. The account here rests on the union’s proposal text, the district’s February 7 bargaining update reporting that it had accepted the union’s proposal on artificial intelligence, and press reporting. Sacramento rests on the California Teachers Association’s own account, which dates the settlement only to “late June” and gives no year; no date is asserted here.
Four Los Angeles documents are in circulation and they do not say the same thing. The district’s April 1 package proposal put the restriction in a memorandum reaching generative AI, set to expire June 30, 2028. The fact-finder recommended the broader “advanced technology, including but not limited to Artificial Intelligence.” UTLA’s summary of what its members ratified places the protection in Article XIII of the contract itself. This piece quotes only the district’s own description to its own board, so nothing here turns on the difference — but the executed article should be read before anyone cites a section number or an expiration date. The only public copy is a 297-page scan without a text layer.
Unions outside the schools have won provisions of both kinds, and some have won more than consent. The Writers Guild’s 2023 agreement, won after a 148-day strike, provides that AI “can’t write or rewrite literary material,” that AI-generated material is not source material, and that a company “can’t require the writer to use AI software” — a flat prohibition, broader than anything a school district has signed. Forty-three NewsGuild contracts contained AI language as of September 2025, and an arbitrator found in 2025 that Politico had violated its clause by deploying two AI products without the 60 days’ notice the contract required — notice and bargaining, which is effects bargaining, not consent. The 2024–2030 USMX–ILA Master Contract, ratified by 45,000 dockworkers at nearly 99 percent, bars signatories from using “automation or artificial intelligence or quantum computing for the performance of clerical functions.” In each case the party that agreed bears the cost of having agreed: the studios absorbed the shutdown, and cargo that finds one port slow moves to another. A school district’s students cannot be routed anywhere.
Government Code §3547.5 and the county disclosure form, item E: “Identify other major provisions that do not directly affect the district’s costs, such as binding arbitrations, grievance procedures, etc.” The Los Angeles filing appears at page 424 of the board’s June 16 materials. In Washington, the cost estimate a Seattle board member receives is required by the board’s own Policy 6015, not by state law; the state’s requirements run to the budget — RCW 28A.505.040, the four-year budget plan reviewed by the educational service district and filed with the Office of Superintendent of Public Instruction — and to RCW 28A.320.055, which asks that the signed agreement be posted within thirty days.
The law is moving the same direction and reaches past schools. California’s AB 2656 would require every public employer in the state to give a recognized employee organization 45 days’ written notice before developing, purchasing, implementing or requiring the use of generative AI to perform work within a represented classification. California’s SB 951 would require an employer — defined to include the state, cities, counties, special districts, local educational agencies and community college districts — to file a written “technology hiring disruption notice” with the Employment Development Department upon “the permanent ending of hiring or contracting for a particular occupation or position, caused in whole or in substantial part by the employer’s use of AI or other automation,” whether or not anyone in that position remains employed and whether or not the total number of positions falls. The reportable event is not a layoff. It is the productivity gain. In New York, S10025 would require twelve months’ notice before a public employer begins procuring AI, and then bargaining over acquiring and implementing it, not merely over its effects. None of these grants a veto. Each establishes that deploying a tool is a labor event, which is where a demand for consent begins.