AI Writing Scandals Expose a Murky Line Between Acceptable and Unacceptable Use
Steven Rosenbaum's book about how AI shapes reality contains more than half a dozen fake or misattributed quotes. The New York Times reported the problem this week. Rosenbaum initially claimed responsibility for the errors, then blamed ChatGPT for "fucking up the book."
The real issue isn't that Rosenbaum used AI-he disclosed that in his acknowledgements. He got in trouble because he used it badly, relying on a chatbot to generate quotes without verifying them, a task at which AI is notoriously unreliable. One fabricated quote attributed to tech journalist Kara Swisher made it into the final manuscript.
This week has brought a cascade of similar problems. A Nobel-winning novelist appeared to admit using AI to sharpen story ideas before claiming misunderstanding. An author won the Commonwealth Short Story Prize with work allegedly written by AI. Two other prize winners came under the same scrutiny. The Commonwealth Foundation initially denied any winners used AI, then reversed course and said it was reviewing the allegations.
The underlying problem is harder to solve than rooting out obvious cases. "AI writing" isn't one thing. It exists on a spectrum.
The spectrum problem
Few would argue that typing "Write a 3,000-word literary short story set in Trinidad" into Claude and publishing the output deserves a prize. But what about a writer who uses a chatbot for research? Or who asks AI to suggest the best word for a sentence? Or who trains a custom AI model to draft first versions of stories?
The tech reporter Alex Heath does the last one. He trained a version of Claude to write in his style and generate first drafts, as reported in March. That's more sophisticated than most writers' use of AI, but it's also more transparent than many practices.
Different organizations have drawn different lines. The New York Times allows freelancers only "high-level brainstorming" with AI tools, while encouraging newsroom employees to experiment with it as "a powerful tool." The Authors Guild warns of ethical risks but stops short of banning any specific use.
Banning AI prose while allowing AI research might seem intuitive. It's also unenforceable-there's no reliable way to detect when AI helped shape a story's framing or research direction.
The hidden uses are the real concern
Neuroscientist Tim Requarth raised a sharper question: what if the hidden uses of AI in the writing process are the actual problem?
The risk isn't that AI generates clunky prose full of the word "delve." The risk is that writers outsource the hard work of discovering truth and interpreting the world. When language models trained on dubious sources and controlled by tech companies influence what gets written about and how, their biases seep into the narratives that shape public understanding.
Is using AI to turn a phrase really worse than using it to decide what to write about in the first place?
The real threat isn't bad writing-it's good enough writing
AI writing scandals don't slow down because the technology keeps improving. If the problem were simply that AI writing was bad, its steady improvement would be good news. Instead, the problem is that AI tools are becoming good enough to fool readers and prize committees.
Rosenbaum told me he cursed ChatGPT for the errors in his book. He also said he couldn't imagine giving it up. That dependence-combined with the technology's superficial competence and human tendency to trust it-may pose a greater threat to writing than any single scandal.
As AI tools become more capable and more common, writers face a choice that journalism, publishing, and literary institutions haven't yet resolved: what counts as acceptable use? The answer remains fuzzy.
For writers navigating this uncertainty, understanding AI's capabilities and limitations is essential. AI for Writers Courses and ChatGPT Courses & Certifications can help clarify how these tools work and where they fail-knowledge that's increasingly necessary for anyone working in the field.
Your membership also unlocks:
