In September, the University of Kentucky excused David Durman, a research administrator, from using AI after he argued that handing judgment and listening over to a machine went against his humanist beliefs. A software engineer who is a Unitarian Universalist won a similar exemption in May, citing the environmental cost of AI. Both used the same federal law, and that law is the real story, because it decides how far this can go.
The employer cannot argue with the belief
Title VII of the Civil Rights Act requires American employers to make room for an employee's sincere religious beliefs, and a belief system without a god, such as humanism, can count. The employer does not get to judge whether the belief is correct. The coal company with the hand scanner learned that: it had treated one worker's belief as a mistake, and a court said that was not its decision to make.
What the employer can do is argue cost. Since a unanimous Supreme Court ruling in June 2023, it has to show that the exemption would cause substantial increased costs for its business. Before that, a small cost was enough to refuse. So the whole contest over an AI opt-out comes down to one question: what does the business lose if this person works without AI?
The more a job is built on AI, the weaker the opt-out
For Durman, the answer is close to nothing. His work was done before AI arrived, and meeting notes and drafting help are things he can do himself.
Now take an insurance claims team where AI drafts the summary of every claim file in two minutes and a person checks it in five. Written by hand, the same summary takes about 25 minutes. An adjuster who handles 15 files a day would need roughly four and a half more hours each day without AI. That is a substantial cost, and an employer in that position has a real case. It could also simply move that adjuster to other work.
The same worker with the same belief gets a different answer, depending on how much of the job has been rebuilt around the tool. Every company that writes AI into job descriptions and sets targets for how much work it should do is building the evidence it would use to refuse an opt-out. That evidence only counts if it exists on paper before the request arrives.
What this means for workers and managers
For an employee in the United States, the route is narrow. The objection has to be religious, not a general dislike of AI, a worry about jobs or a political view. The belief has to be deeply held, and you have to be able to show it. The American Humanist Association has invited humanists to ask it for legal help with requests like Durman's. Outside the United States, this law does not apply, and other countries have their own rules.
For a manager, the useful work is done before anyone asks. For each role, write down which tasks truly need AI and what they cost without it. Where the answer is "not much", an exemption is easy to grant. Where the answer is hours a day, you will have a number to show.
An exemption request makes a company say, role by role, what AI is actually doing for it. A company that cannot answer will struggle to refuse the request, and it probably does not know what its AI rollout is worth either.