“AI is unethical, unless it helps me.”
That, increasingly, seems to be the guiding principle of the AI era. Not artificial intelligence. Not artificial general intelligence. Just… artificial exceptions.
There’s a growing tendency to apply one ethical standard when AI threatens our profession, our company or our livelihood, and a completely different standard when it happens to benefit us.
Call it AI Exceptionalism.
It’s the belief that AI should follow different rules depending on who is using it.
To be clear, this isn’t an argument that all uses of AI are equally good. Nor is it an argument that copyright doesn’t matter, or that artists, writers, actors and educators don’t have legitimate concerns.
They absolutely do.
But if we’re going to have an honest conversation about AI, we should probably try applying the same principles consistently.
Here are four examples where AI consistency disappears remarkably quickly.
On this page
- AI shouldn’t write articles, but it can absolutely write code
- Training on copyrighted books is fair use, but training on our AI is theft
- AI shouldn’t replace actors, but replacing everyone else seems negotiable
- Students shouldn’t use AI, but universities definitely should
- The Uncomfortable Question
AI shouldn’t write articles, but it can absolutely write code
Some of the strongest criticism of generative AI has come from journalists themselves.
Take Kara Swisher. She has repeatedly warned about generative AI’s impact on journalism, describing concerns around misinformation, media quality and the erosion of trust. Yet she’s also spoken enthusiastically about AI’s potential in software development and has described AI coding tools as one of the genuinely transformative applications of the technology.
Or consider Kevin Roose of The New York Times. Roose has written extensively about the dangers AI poses to journalism and the creative industries, while also documenting his own experiences using AI programming assistants and exploring how dramatically they can improve developer productivity.
Technology journalist Casey Newton has similarly argued that AI presents profound challenges for writers and publishers, while frequently highlighting rapid advances in AI-assisted software engineering and discussing the remarkable capabilities of coding models.
None of these journalists are necessarily contradicting themselves. They may genuinely believe there are important differences between writing news articles and writing software. But those differences are rarely articulated.
Instead, an interesting pattern emerges.
- AI writing is often discussed as replacing skilled creative professionals.
- AI coding is often discussed as augmenting skilled creative professionals.
Which raises an awkward question:
Why is writing code fundamentally different from writing prose?
Both require creativity, experience, judgement and years of practice. Both are professional crafts. Both are capable of being partially automated. If AI assistance is acceptable because it makes programmers more productive, why isn’t the same argument valid for journalists?
Or, if AI-generated writing undermines professional creativity, shouldn’t AI-generated code raise exactly the same concerns?
Perhaps there is a meaningful distinction. If so, it’s a conversation worth having explicitly.
Because otherwise, to an outside observer, it can look suspiciously like AI is judged less by what it does and more by whose profession it affects.
Training on copyrighted books is fair use, but training on our AI is theft
This might be the clearest example of AI Exceptionalism currently playing out.
In early 2025, OpenAI publicly accused DeepSeek of improperly distilling OpenAI models into competing systems.
Anthropic has since made similar allegations against DeepSeek, Alibaba and other Chinese AI labs, claiming they created fake accounts to extract Claude’s behaviour at scale.
Both companies argue that model distillation unfairly copies years of research and billions of dollars of investment.
That’s a perfectly understandable position. Except…
Those same companies continue to argue in court that training frontier AI models on enormous collections of copyrighted books, newspapers, websites, photographs and source code is legally permissible — often under the doctrine of fair use. Those arguments sit at the heart of ongoing copyright lawsuits brought by authors, publishers and creators.
So we end up with two remarkably similar statements.
- Learning from millions of human-created works without permission? Innovation.
- Learning from an AI model without permission? Theft.
Of course, OpenAI and Anthropic argue there are important legal and technical differences. Model weights are proprietary. Distillation reproduces unique capabilities. Training data is transformed rather than copied.
Those arguments may ultimately succeed in court. But to many observers, the optics are difficult to ignore. When someone learns from your work, it’s innovation.
When someone learns from my work, call the lawyers.
AI shouldn’t replace actors, but replacing everyone else seems negotiable
The 2023 Hollywood strikes brought AI into the mainstream. Actors raised genuine concerns about digital doubles, voice cloning and synthetic performances, eventually leading to new protections in the SAG-AFTRA agreement covering consent and compensation for AI replicas.
Those concerns weren’t hypothetical. Studios were actively exploring them. That’s a good thing. But something else went largely unmentioned.
Hollywood has enthusiastically embraced AI everywhere else.
AI-assisted rotoscoping, de-aging, background replacement, facial cleanup, animation workflows and visual effects are increasingly reducing work traditionally performed by VFX artists, compositors and junior production staff.
The ethical question becomes interesting.
If AI replacing actors is unacceptable because it threatens creative workers, why isn’t the same concern raised with equal force for everyone else working on the film?
To be fair, SAG-AFTRA exists to represent actors - not visual effects artists. That’s literally its job. But from the outside, the broader ethical principle can start to look selective.
“My profession deserves protection.”
“Yours is… more complicated.”
Students shouldn’t use AI, but universities definitely should
This one may feel familiar to anyone currently studying.
Many universities now prohibit or heavily restrict students from using ChatGPT, Claude or Gemini in assessments while simultaneously publishing AI strategies encouraging staff to embrace generative AI for administration, teaching support and operational efficiency.
Universities including Oxford, Cambridge, Harvard and countless others have all published guidance drawing this distinction in different ways.
The reasoning is understandable. Universities exist to assess learning. If AI completes the assignment, what exactly is being evaluated?
But walk a little further across campus. Universities increasingly use AI themselves.
Admissions. Student support. Administrative automation. Research assistance. Marking support. Academic integrity systems. Email drafting. Scheduling. Operational planning.
In other words:
- Students are told AI undermines learning.
- Staff are told AI improves productivity.
Again, perhaps there really is a meaningful distinction. Assessment and administration aren’t the same thing.
But students naturally ask the obvious question.
If AI makes professors more productive, why shouldn’t it help students become more productive too?
The answer might be “because learning matters.” That’s a perfectly reasonable position.
But it’s a conversation worth having openly — not one resolved by simply declaring one use virtuous and the other forbidden.
The Uncomfortable Question
Notice something? Every example follows the same pattern.
- When AI threatens my profession, it’s unethical.
- When AI benefits my profession, it’s innovation.
It’s remarkably human. We all do it.
Doctors worry about AI diagnosing patients. Lawyers worry about AI drafting contracts. Programmers worry about AI replacing developers. Designers worry about AI generating artwork. Journalists worry about AI writing articles.
Everyone becomes an AI ethicist the moment AI starts competing with their job.
Maybe the real conversation isn’t about AI. Perhaps it’s about incentives. Humans have always been remarkably good at discovering ethical principles that just happen to align with their own interests.
AI hasn’t changed that. It’s simply made it much easier to notice.
Because whether you’re a journalist, a Hollywood studio, an AI company or a university, the exceptions tend to appear at exactly the point where AI becomes personally useful.
So what should the rule be? Should AI be allowed everywhere? Restricted everywhere? Licensed? Compensated? Transparent?
Perhaps.
But whatever principles we choose, they should probably apply consistently. Otherwise we aren’t really debating AI ethics.
We’re debating who gets to benefit from AI — and who doesn’t.
That’s not AI ethics. That’s AI Exceptionalism.
And if there’s one thing humans have always excelled at, it’s believing that the rules are different when they apply to us.
