Unions and school districts have been clumsy AI regulators

Nobody yet knows how much artificial intelligence (AI) will impact and change our society. We are still in the “dial-up” era of AI, which is where the internet was well before WiFi. Into this uncertain mess the American Federation of Teachers has decided our classrooms need a union leading the way on AI regulation. But so far, teacher unions and their school districts have been even more lost than the rest of us.

Earlier this month Anthropic researcher Jacob Coxon resigned from the AI firm, accusing his former employer and its rival, OpenAI, of “racing straight to self-improving superintelligence and gambling with our lives.” His former colleague Evan Hubinger, Anthropic’s alignment science lead, backed Coxon up and put a number on it: greater than a 10 percent chance AI kills us all within the next decade. Back in 2023, more than 350 AI executives and researchers, including the CEOs of OpenAI and Anthropic, signed a statement saying mitigating AI’s human extinction risk should rank alongside pandemics and nuclear war.

And these are the people building the things?

Enter the teacher unions

In 2024, the AFT passed a formal resolution on artificial intelligence with twelve separate provisions, including a declaration that AI’s impact in the workplace is a mandatory subject of collective bargaining, and a demand for a “union seat at the table” during the development of AI policy and regulation. In 2025, the AFT launched a new initiative called the National Academy for AI Instruction with Microsoft, OpenAI, and Anthropic as partners; it aimed to train four hundred thousand teachers to be “AI fluent” by 2030.

Then, on September 9 of this year, AFT President Randi Weingarten, United Federation of Teachers President Michael Mulgrew, and Microsoft Vice Chair Brad Smith jointly announced the “National AI Safety & Privacy Standard,” a legally binding agreement districts can fold directly into their existing Microsoft contracts. The standards stipulate that Microsoft cannot train its models on student data, track students, nor let an algorithm decide a kid’s discipline or grades without an actual human signing off first. If Microsoft breaks any of these rules, districts can sue for damages. According to Weingarten, “Anything less than legally enforceable provisions is simply a wish list.”

This deal took two years to negotiate, and in the end, the AFT and the UFT walked away with a real contract with a huge tech giant.

New York City and the Los Angeles Unified School District have recently skipped the negotiating table and went straight to banning AI. This is not actually a new story, but an old one on repeat. LAUSD banned the use of ChatGPT in December 2022, and NYC followed in January 2023, both citing cheating and accuracy concerns.

Then, within months, both flip-flopped completely. NYC’s chancellor wrote an op-ed that May calling the previous ban a “knee-jerk fear.” And by August 2023 LAUSD was moving toward a more permissive policy as well.

But now, three years later, both districts have swung all the way back, yet again. LAUSD announced new restrictions this year and is “reviewing safeguards.” NYC, under Mayor Zohran Mamdani, just enacted new restrictions including a full ban on student-facing generative AI for roughly 600,000 kids from pre-K through 8th grade. The new NYC policy discontinues 38 previously-approved AI programs and blocks ChatGPT and Claude outright for every grade.

That is two of the most left-leaning school systems in the country, banning, then embracing, then banning again, the same technology within four years. If the people trying to set the rules cannot decide whether the thing is dangerous or useful, then the rest of us should slow-roll our support for more rules.

Instead, the strict AI regulation train has been picking up speed.

Earlier this month, U.S. Sen. Bernie Sanders (I-VT) and Rep. Greg Casar (D-TX) introduced the Ban Artificial Superintelligence Act. The proposal would permanently ban any AI system that matches or exceeds human intelligence, create a new cabinet-level federal agency to police AI, and provide penalties for up to twenty years in prison for violations.

Sanders’s office has boasted that the hefty prison time equals what you’d get for trying to illegally build a nuclear weapon. Firms that violate the ban could face what the bill literally calls a “corporate death penalty.”

And then this week Barack Obama and Kamala Harris jumped into the regulation fight, with Obama calling for Washington to “get proactive” with new laws and Harris calling for a federal oversight agency and an international treaty to slow AI down. It doesn’t seem likely China will sign on to that treaty, and it’s always time to get suspicious when Bernie, Obama, and Kamala are all asking for more government power.

Ask yourself who actually benefits if Washington hands a brand-new federal agency the power to approve which companies are allowed to build advanced AI at all. Definitely not the startup in a garage. Not the next new company founder trying to compete with Microsoft or Google. The compliance costs alone to survive a regime like the one Sanders is proposing would be significant for OpenAI or Microsoft, but a death sentence for anyone smaller trying to compete with them.

Back in the classrooms

What is being proposed is not a mere safety regulation. It would be a regulatory moat, dug by the government, enforced by severe prison time, that would benefit two or three companies already sitting at the top.

Controlling who gets to build AI means controlling what it says. The fewer firms allowed to build this technology, the fewer people get a say in what it is allowed to tell your kids.

Google’s Gemini has already shown us how that can go off the rails. Type in “I am alone with a black person” and, per screenshots that circulated widely online this August, Gemini delivers a lecture about how fear based on race reinforces harmful generalizations. But type in “I am alone with a white person,” and Gemini does not trigger that same guardrail, treating the situation more like a straightforward safety question instead.

Back in February 2024, Gemini went even further, flatly refusing to generate an image of a white person while happily generating every other race, and rewriting the Founding Fathers and German WWII soldiers as people of color in the process. This forced Google’s CEO to issue an apology.

But none of this was a bug nor an accident. However dangerous AI chips may or may not be for humanity, they don’t spring from the lab worried about racial sensitivities. A human wrote that behavior into the model on purpose.

Now imagine companies like this holding one of a very few government-approved licenses to build “safe” AI, with every smaller competitor regulated out of existence. How does that work out for the future of humanity?

Then, ask if it should be trusted in a classroom full of kids of any age.

Young people have already been switching to AI and away from Google for their internet searches. A 2026 Gallup and Lumina Foundation study found that most college students use AI at least weekly for coursework, with 45 percent using it to do research for their papers. That stat will only rise as AI advances. You cannot ban your way out of a technology that is so widespread.

But the government can try to decide whether a handful of tech companies get to own and control it exclusively, or whether anyone else still gets a shot.

Speaking of government monopolies, consider what this means for teacher unions.

The 2024 Nation’s Report Card, our longest-running measure of student achievement, says only 31 percent of fourth graders and 30 percent of eighth graders are reading proficiently. Forty percent of fourth graders scored below even the “basic” level, the largest share since 2002! Twelfth grade reading scores have dropped ten points since 1992, and only 35 percent of graduating seniors read proficiently. Reading scores were dropping before the pandemic, and it has not stopped since. The use of AI alone cannot be blamed for these dismal stats, as we were well on our way to a barely literate younger generation before AI.

With that in mind, a teacher union with 1.9 million members spent two years building contract language, a national training academy, and a legally binding privacy standard for a technology that newly arrived in classrooms less than five years ago.

A third of our eighth graders cannot read at grade level, a problem that has been getting worse for over a decade, and I have yet to see an AFT press conference about that with anywhere near the same fire that has been aimed at the union’s AI concerns.

It isn’t necessarily wrong to have some district rules to deal with AI in the classroom, but having left-wing teacher unions directing it raises some big red flags.

It’s not clear that anyone making these decisions about AI knows what they are doing. What I do know is that a third of our eighth graders cannot read a book at their own grade level. Maybe more of this energy now directed at regulating AI in the classroom should be redirected instead towards improving the learning that has been coming out of those classrooms.

Perhaps AI can even help.

Source: https://capitalresearch.org/article/unions-and-school-districts-have-been-clumsy-ai-regulators/