David Durman’s AI Exemption Challenges the Assumption That Everyone Must Adopt It

David Durman secured permission to avoid personally using AI at work, despite grounding his request in beliefs that require no god. David Durman’s AI exemption, approved by the University of Kentucky, turns an increasingly familiar workplace expectation into a question about freedom of conscience.

The research administrator sought protection before suggestions to adopt AI became a requirement. His employer accepted an arrangement that allowed him to continue performing his job. The result challenges the assumption that adopting AI must become part of every employee’s professional identity.

The dispute sits between organizational pressure to use AI and an individual’s commitment to exercising human judgment. Shopify’s public embrace of AI as a baseline expectation provides a useful contrast. Kentucky’s decision shows another approach: assess whether an employee can deliver the required work while following a different process.

David Durman’s AI exemption arrived before an AI mandate

The university approved an individual accommodation without requiring Durman to abandon his existing responsibilities.

The American Humanist Association announced the personal AI accommodation on October 26. According to the organization, Durman submitted his request in late August and met university officials in mid-September. Approval followed on September 25.

The association helped him pursue the request through the university’s Office of Equal Opportunity. It says he continues performing all his duties without personally using AI.

That sequence matters because the public announcement followed the actual decision. The October news cycle concerns an accommodation already approved in September, rather than a newly issued workplace ruling.

It also matters because this was an employer’s decision about an employee. It was not a court judgment establishing a nationwide exemption from AI requirements.

According to TechCrunch’s reporting on the AI exemption request, Durman had not been required to use AI. Colleagues and leadership had recommended trying the technology. He sought protection before those suggestions developed into a mandate.

University spokesperson Jay Blanton told the publication that Kentucky considers requests individually, taking account of responsibilities and circumstances. He said the arrangement allowed Durman to continue performing his position’s essential functions. He also indicated that accommodations can receive further review as technology and duties change.

Those details narrow the story considerably. There is no established confrontation here between a university demanding automated work and an employee refusing an assigned task.

Instead, Durman sought clarity while his existing working methods remained acceptable. The university recognized that his beliefs and his responsibilities could coexist.

For employers, that distinction creates a practical question before any legal confrontation begins. Does the organization need a particular result, or does it specifically need every employee to use a particular method?

Those requirements can overlap, but they should not automatically become interchangeable. An administrator might complete an assignment through direct reading, personal correspondence, and independent drafting. Another employee might use an AI assistant while pursuing the same outcome.

Whether both approaches remain workable depends on the assignment. Durman’s case establishes that Kentucky accepted the distinction for his position.

The timing also changes the meaning of his request. It addressed an emerging expectation before a disciplinary dispute forced everyone into opposing positions.

That does not establish that early requests always succeed. It shows that a workplace can resolve a potential conflict through an accommodation process while the employee continues working.

The broader significance lies in that ordinary administrative result. A growing technology expectation encountered an existing employment process, and the institution found room for an exception.

Employers face a question about methods, not just results

An AI adoption strategy becomes harder to defend when it treats tool usage as a substitute for examining actual work.

For managers, the immediate pressure is operational. They must identify what an employee needs to accomplish and which parts of the process genuinely require AI.

Consider a hypothetical research office preparing a grant submission. An AI assistant might help summarize instructions, suggest language, or organize background material. A person might also complete those activities by reading the documents and writing directly.

The relevant comparison would include accuracy, deadlines, confidentiality, and coordination. Counting prompts alone would reveal little about whether the office completed its responsibilities.

This example illustrates the management question rather than documenting Durman’s specific workflow. The publicly described accommodation does not provide a detailed account of his assignments or performance measures.

Still, the distinction between adopting software and delivering work is central to understanding the case. An employer can encourage experimentation without making enthusiasm itself the measure of competence.

The difficulty increases when an organization builds AI into shared processes. A document assistant might be optional for an individual writer but embedded in a team’s review system.

That creates several separate decisions. Who generates the material? Who checks it? Who submits it? Who remains accountable when something goes wrong?

An accommodation becomes easier to assess when those responsibilities remain visible. A vague instruction to “use AI more” obscures the actual conflict.

It also makes reasonable disagreement harder to resolve. An employee might object to generating text while accepting an ordinary document portal. Another might object to automated recording or analysis.

Those differences should not be collapsed into a single category of resistance. They concern different actions and potentially different workplace arrangements.

The same analysis matters to enterprise buyers selecting AI products. A purchasing decision can affect employees who never choose to open a chatbot.

If AI features become part of standard communication systems, organizations need to understand which functions remain optional. Product documentation about controls becomes relevant to workplace policy.

A hypothetical buyer evaluating meeting software, for example, should distinguish basic attendance from automated transcription and summary generation. That distinction identifies what participation actually requires.

It does not guarantee that every feature can be disabled. It makes the tradeoff concrete enough for an institution to evaluate.

For knowledge workers, the story also raises a broader question about professional judgment. Employers often expect people to take responsibility for the quality of their work.

An AI requirement introduces another question: how much freedom should employees retain over the process that produces that work?

Durman’s case does not settle that question across occupations. It makes the question harder to ignore when organizations introduce broad expectations without examining individual duties.

AI workplace religious accommodation therefore belongs in conversations involving managers, human resources, legal teams, and technology buyers. Treating it exclusively as a software training issue misses the conflict.

The pressure falls on organizations to describe their needs precisely. Clear requirements provide a stronger foundation for both adoption and accommodation than a general demand for participation.

Religious protection does not require belief in a god

The legal category of religion can encompass qualifying nontheistic beliefs, but strong opposition to AI does not automatically qualify.

Title VII, the federal employment discrimination law, supplies the legal framework behind the reported accommodation. Its treatment of religion extends beyond membership in a traditional faith.

The Equal Employment Opportunity Commission’s religious belief guidance recognizes sincerely held nontheistic moral or ethical beliefs with a place comparable to traditional religious convictions.

The agency also distinguishes qualifying beliefs from personal preferences and social, political, or economic philosophies. Intensity alone does not turn an objection into a protected religious belief.

This distinction explains why the headline’s reference to belief in a god matters. The absence of theism does not itself defeat a request. Conversely, describing an objection as deeply felt does not itself establish entitlement.

The resulting inquiry concerns the individual’s beliefs and their relationship to the disputed activity. It does not require every member of a tradition to reach the same conclusion.

For Durman, the reported conflict concerns the exercise of human judgment. Humanism, a worldview emphasizing human welfare and responsibility, informs his objection to delegating thinking to AI.

The American Humanist Association also emphasizes that humanists hold different views about responding to AI. Durman’s position should therefore remain his position, rather than becoming a supposed universal humanist prohibition.

That distinction protects the integrity of the story. An individual accommodation does not establish that an entire belief community rejects the technology.

It also keeps the analysis focused on conscience rather than theological classification. The relevant workplace question concerns the claimed conflict and an arrangement that addresses it.

A useful comparison involves two hypothetical employees declining the same writing assistant. One prefers drafting independently because the software produces awkward prose. Another describes its use as conflicting with a comprehensive belief system.

Their outward conduct looks similar, but the underlying reasons differ. That difference helps explain why a general workplace preference and a religious accommodation request require separate analysis.

Neither example predicts the outcome of an actual case. The employee’s circumstances and the employer’s responsibilities remain relevant.

There is also an important distinction between recognizing a belief and endorsing its factual claims. An employer can accommodate someone without agreeing with every argument they make about AI.

Durman’s concerns about human judgment do not establish that every AI interaction diminishes moral responsibility. His successful request does not resolve research questions about cognition or automation.

Those questions deserve their own evidence. A workplace accommodation evaluates a different relationship: the employee, the conflict, and the practical arrangement.

Seen this way, the Durman humanist accommodation explained through employment law is less mysterious than the headline initially suggests. Existing protections encountered a new category of workplace tool.

The surprising element is the application, rather than the discovery of an entirely new legal principle. AI has entered a familiar area where institutional expectations meet individual convictions.

AI-first workplaces create the opposing pressure

The central conflict concerns whether universal adoption is necessary for an organization to achieve its goals.

Shopify offers a concrete comparison because it has publicly described AI use as an organizational expectation.

In its account of baseline AI usage, the company recalls Tobi Lütke’s April 2025 memo establishing reflexive adoption as the default. The company presents experimentation and shared learning as ways to improve its work.

That position reflects a recognizable management argument. Employees who never experiment cannot discover where a technology helps, where it fails, or how colleagues are using it.

An organization can also struggle to build shared practices when everyone chooses unrelated tools. Common systems can simplify collaboration, support, and review.

Those are reasons to take adoption seriously. They do not establish that every employee must personally use every AI capability to contribute effectively.

Kentucky and Shopify are different organizations with different responsibilities. Comparing their public positions does not establish how Shopify would handle a request resembling Durman’s.

Nor does this case show that a university’s accommodation would transfer unchanged to a technology company. The comparison concerns management approaches, rather than identical legal facts.

One approach starts with a broad expectation of experimentation. The other, in this instance, accepted an individual exception while preserving the employee’s work.

The tension becomes sharper when AI use serves as a cultural signal. Managers might interpret participation as curiosity, flexibility, or willingness to learn.

But a tool usage measure cannot explain why someone declines a tool. It cannot distinguish conscientious objection from inadequate training, poor product fit, or simple preference.

That ambiguity gives employers a reason to examine performance and process separately. Otherwise, adoption metrics risk becoming an indirect assessment of agreement with management’s technology philosophy.

There is a legitimate counterargument. Allowing different methods can complicate collaboration, especially when a shared process depends on consistent inputs.

A hypothetical team might need standardized records for review or handoffs. An alternative workflow that produces incompatible material would create a different problem from independent drafting alone.

The appropriate analysis would examine that concrete incompatibility. A general statement that everyone should embrace AI would leave the underlying operational issue unexplained.

Product teams also have a stake in this distinction. A system offering clear feature controls gives customers more choices when accommodating different working methods.

That observation does not prove that optional controls solve every dispute. It identifies a design consideration that becomes more valuable when organizations need precise exceptions.

The same principle applies to personal knowledge management: retaining and organizing information is a goal that can involve different methods.

Organizations should be able to identify where AI contributes to that goal and where human work remains an acceptable alternative.

David Durman’s AI exemption brings this distinction into public view. It challenges an adoption assumption without establishing that AI-first organizations must abandon their strategies.

The exemption has boundaries, and its broader reach remains uncertain

Permission to avoid personal AI use does not necessarily remove an employee from every workplace process involving AI.

Religion News Service’s reporting, republished by The Christian Century, describes the accommodation’s practical limits. Durman remains expected to participate in meetings or submit materials through systems where colleagues use AI.

That boundary matters because personal use and organizational exposure are different things. An employee can refrain from prompting an assistant while still working alongside people who use one.

The arrangement also illustrates a possible compromise. An institution can preserve shared participation while changing what an individual must personally do.

It does not follow that every employee would find that boundary consistent with their beliefs. A different objection might concern recording, processing, or other people’s handling of submitted material.

Those would present different facts. Durman’s accommodation cannot supply a predetermined answer to every dispute involving AI somewhere in a workflow.

The employer’s burden matters as well. The EEOC’s current accommodation standard explains that the Supreme Court’s 2023 Groff v. DeJoy decision rejected a minimal-cost understanding of undue hardship.

The standard considers whether a burden is substantial in the overall context of the employer’s business. Relevant circumstances include the accommodation’s practical impact and the employer’s operations.

This framework does not make accommodation automatic. It requires more than treating a minor inconvenience as sufficient grounds for rejection.

For readers assessing this story, the distinction prevents two opposite exaggerations. The decision does not establish an unrestricted right to refuse AI, and managerial preference does not explain every possible denial.

Publicly available reporting also leaves practical questions unanswered. It does not provide enough information to calculate the arrangement’s long-term effects on deadlines, staffing, or departmental performance.

There is no basis for claiming the exemption improved productivity. There is equally no basis for claiming that it undermined the university’s work.

The university’s stated willingness to review accommodations as circumstances change adds another limit. An arrangement workable under current responsibilities might require discussion if those responsibilities change.

That possibility should not be confused with evidence that review is already underway. It describes the institution’s position about future circumstances.

David Durman’s AI exemption therefore carries more weight as a documented example than as a forecast of national outcomes. It shows that an employer accepted this arrangement.

The strongest skeptical question concerns transferability. What happens when the disputed technology sits closer to the central task of a position?

A hypothetical employee hired specifically to evaluate AI systems would present different operational questions from an administrator using established methods. That example illustrates why job design matters.

It does not determine whether a particular request succeeds. Any such conclusion would require facts beyond this case.

The article’s central tension survives those limits. Organizations seeking broad AI adoption still need to explain when personal use is essential and when another method remains workable.

Watch the next decisions, the workflows, and the evidence

The most useful developments will show whether organizations can preserve both meaningful work requirements and workable individual accommodations.

The first signal over the coming months is additional documented decisions. Written approvals, denials, or publicly described disputes would reveal whether comparable requests receive similar treatment.

Specificity will matter more than headline volume. A request’s outcome means little without knowing the employee’s duties, the objection, and the arrangement considered.

Comparable approvals would strengthen the view that individualized exceptions can coexist with adoption strategies. Disputes grounded in different job requirements would clarify where that approach encounters limits.

The second signal is how employers describe mandatory AI use. Policies that identify particular tasks would provide more useful evidence than broad encouragement to experiment.

For example, a hypothetical policy might require a shared submission format while leaving drafting methods open. Another might require direct interaction with an AI system.

Those requirements create different conflicts. Watching how organizations distinguish them will reveal whether adoption policies become more precise.

Clear alternatives would support the argument that employers can separate results from methods. Policies tying essential duties closely to AI would make that separation harder.

The third signal is evidence about how accommodations function after approval. Useful information would concern completed work, coordination, deadlines, and any changes to responsibilities.

Neither enthusiastic announcements nor predictions of disruption can substitute for that record. The practical outcome depends on whether the arrangement remains workable over time.

This is also where the debate needs restraint. A successful accommodation would not prove that AI lacks value. A difficult implementation would not invalidate every conscientious objection.

Each outcome would illuminate a particular relationship between a job, a person, and a process.

For employees following the story, David Durman’s AI exemption provides a concrete example to examine rather than a universal permission slip. For employers, it provides a reason to define expectations before turning them into performance requirements.

Read the underlying announcement and the official guidance with those distinctions in mind. Ask which activity creates the conflict, which outcome the organization needs, and what evidence establishes the practical burden. The next useful question is whether a workplace can describe those requirements clearly enough to assess an alternative fairly. As AI enters more ordinary tasks, that clarity will matter to employees who embrace it and those who seek another way to work.