You can refuse to use AI at work on religious grounds
A University of Kentucky administrator just became the first Humanist in the US to win a formal exemption from using AI on the job. Two cases in six months is no longer an anecdote — it is a policy.
David Durman, a research administrator at the University of Kentucky, requested a religious accommodation in late August, met with the university's Office of Equal Opportunity in mid-September, and had it approved on September 25. He asked to be excused from personally using AI on the basis of his "sincerely held beliefs as a Humanist," and the office concluded that granting it posed no undue burden on his department, according to the American Humanist Association, which helped draft the request. Durman, whose job is supporting researchers applying for grants, still performs every part of his duties. He just performs them without a model.
That is the entire news, and it is easy to file away as a curiosity about one employee's conscience. It isn't. What Durman received was not a favour from a sympathetic manager. It was a legal accommodation under Title VII of the Civil Rights Act, processed by the same office that handles race and disability complaints, decided on the record, with a date attached. The precedent is the paperwork, not the man.
Why a religious claim, and not an ethical one
Durman's stated objections are the ones AI critics make constantly: that the technology interferes with people's ability to earn a living, that its environmental costs are severe, and that it takes from human creativity. None of those arguments have any legal standing on their own. There is no conscientious-objection right for software in American employment law, and an employee who simply dislikes the tools has no protection at all.
Religion is the one channel that exists. Title VII requires an employer to accommodate an employee's religious observance and practice unless doing so would cause undue hardship, and it defines religion broadly enough that courts have spent decades arguing about where the line sits. The AHA's case is that the line comfortably includes a non-theistic moral system. "Comprehensive non-theistic belief systems like Humanism are entitled to the same religious protection under Title VII and the United States Constitution as 'traditional' religions," Alexandria Huiskens, a staff attorney at the association, said in its announcement.
The bar for refusing such a request also got much higher three years ago. In Groff v. DeJoy, a unanimous Supreme Court held that an employer denying a religious accommodation must show the burden would produce "substantial increased costs in relation to the conduct of its particular business," per the opinion — replacing a decades-old reading that let employers refuse once a cost climbed above the trivial. "It's inconvenient" and "it costs a bit" stopped being defences in 2023. An employer that wants to turn down an AI exemption now has to document what the refusal actually costs in rework, coverage and delay, and that is a much harder memo to write than a shrug.
The case that made this copyable
Durman is the second person to do this publicly, and the first whose claim rests on a secular belief system. Erin Maus, a 34-year-old software engineer at a tech entertainment company in North Carolina, requested a religious accommodation in April as a Unitarian Universalist and had it granted in mid-May, as Futurism reported, citing Business Insider's interview with her. "I'm writing my code and reviewing my code by hand, which seems crazy to say," Maus said. "Just two years ago, how else would you do it?"
Her case is the harder one to argue against, because a working engineer declining AI assistance is a genuine operational cost, and she reported that her hand-written work came in about as fast as a colleague's AI-assisted work. Durman's is the easier one: an administrator who declines a chatbot is mostly declining a habit, which is precisely why the university could grant it without strain. Two different cost profiles, one shared legal mechanism — and the mechanism is what other employees will now copy. A public letter, a meeting with the equal-opportunity office, a moral framework invoked by name. The association has published the write-up and a workplace page, so the template is available to anyone who wants it.
What the employer actually agreed to
The University of Kentucky's account is worth reading closely, because it is the template every HR department will end up working from. Spokesperson Jay Blanton told TechCrunch that the institution reviews accommodation requests individually, weighing the employee's responsibilities and the circumstances of the request, and that "the university determined that an accommodation could be provided while allowing the employee to continue performing the essential functions of his position," TechCrunch reported.
The same statement carries the escape hatch: "As technology and job responsibilities evolve, accommodations may be reviewed to ensure they remain reasonable and workable." Treat that as a lease with a renewal clause. The grant is not permanent, and the trigger for review is the employer's own judgment about what the job has become. If AI use migrates from recommended to required in a role, the analysis changes with it — and nothing in the September decision prevents the university from saying so later.
It is also narrower than it sounds. The accommodation covers Durman's personal use of AI, which means the department can still run on AI-generated work; whether he can be handed a machine-drafted document to review is a separate question his letter does not appear to settle. Employees reading this as a right to never touch an AI artifact should read the wording again.
The three objections that will decide the next case
Sincerity. Courts do not judge whether a belief is true, but they do judge whether it is sincerely held and genuinely religious rather than a preference dressed up in moral language. A worker who finds the tools distasteful, or simply less reliable, is not protected. A worker whose objection sits inside a framework that predates the technology — Humanism, Unitarian Universalism, a Catholic social teaching — is. That distinction is doing an enormous amount of work, and it is exactly what a refused applicant will litigate.
Essential functions. If operating AI systems is an essential function of the role, the accommodation is not an accommodation but a redesign, and employers can refuse. This is why a grant-support post was grantable and why Maus's engineering post was unusual: most software jobs now assume the tools.
Coworker burden. Groff did not delete hardship from the test, it raised the evidence bar. Impact on colleagues counts. In a five-person team where one person declines AI, the work lands on the other four, and that shift is the employer's strongest exhibit if it wants to say no.
There is a fourth, less legal objection: that this becomes a loophole for anyone who would rather not learn the tools. That worry is real but misplaced — the sincerity test exists to catch it, and a court that has watched religious accommodation claims for fifty years is not easily fooled by a sudden conversion to Humanism the week a mandate lands. What the cases actually establish is narrower and more useful: an employee with a coherent moral objection and a role that can be done without AI has a strong claim, and an employer that refuses without a cost analysis is exposed.
What to watch
Three things. Whether a refused request produces the first suit testing AI use as a religious burden — that is the case that will define the doctrine, and the plaintiff will need an employer careless enough to refuse without paperwork. Whether professional bodies formalize a position; the Unitarian Universalist Association had no AI policy when Maus won and was drafting one, and a formal statement from a denomination makes the next claim easier to certify. And whether any state adds AI-mandate language to its labor code, which would move the argument from accommodation to statute.
For employers, the practical sequence is short: route a written refusal through the accommodation process rather than to the manager who assigned the tool, put the essential functions of the role on paper before the meeting, and decide in writing with a review date. If you refuse, keep the cost evidence. If you grant, keep the review clause.
We wrote earlier this year about the other half of this dynamic — Silicon Valley has started punishing the AI doomers: workers who voice objections to the tools tend to pay for it. The interesting thing about the Durman and Maus cases is that the objection finally has a form that costs the employer something to ignore.
Would you ask your employer for an AI exemption, or is the request itself the risk? Write to us via our contact page.
