Most of my school years were spent learning in a language I was still learning.
Sometimes I understood the lesson but missed the joke. Sometimes I knew the answer but could not find the words quickly enough. By the time I had a voice I could call my own, it had already been shaped by five languages and half a dozen countries.
So when someone asks me to prove that a sentence is really mine, I understand why they are asking. I am just not sure what kind of proof would satisfy them.
This year, the Commonwealth Short Story Prize had to ask a version of that question at scale. After several regional winners were accused of using generative AI, the Commonwealth Foundation asked them for earlier drafts, notes, photographs, and explanations of how their stories had come together.
In other words: SHOW US THE TRAIL.
The writers were eventually cleared. But the request itself points toward a strange future. A writer may no longer be able to prove authorship simply by handing over the finished work. The proof may have to sit behind it, in notebooks, abandoned openings, rejected metaphors, source choices, and changes of mind.
I can see why this feels reasonable.
A literary prize has a reputation to protect. It cannot simply ignore a serious allegation. And if a writer can produce the notebook, the early draft, and the pages that did not work, that surely tells us something.
But only something.
A paper trail can show that a draft existed before a certain date. It cannot tell you which language I was thinking in when I wrote it. It cannot explain why one sentence sounds Portuguese, another sounds English, and a third carries the hesitation I learned in Japanese.
What would my “natural” English even be?
The English I learned at school? The English I picked up from novels? The English I used at work in Singapore, Malaysia, Britain, or Hong Kong? The English corrected by teachers and editors over the years?
And what would count as my natural Japanese, when it carries traces of Portuguese and English, along with the years I spent learning how another culture places distance, obligation, and hesitation inside a sentence?
For someone like me, fluency was never one stable thing.
It was built slowly and unevenly out of everywhere I had lived.
AI did not create this problem. It arrived in the middle of it.
Long before detectors existed, multilingual and immigrant writers were already being asked why their English sounded the way it did, or why their first language contained gaps that another native speaker’s did not.
The suspicion was already there: perhaps a voice formed between cultures is somehow less authentic than one formed safely inside a single language.
AI simply gives that suspicion a score.
That is what worries me.
Research on authorship attribution across different languages has already shown that methods which appear reliable in one language do not necessarily work across other language families and writing systems.
And institutions do not encounter language under laboratory conditions. They encounter bilingual students, immigrant engineers, accented executives, translated documents, heavily edited prose, and people whose fluency changes depending on the subject, the audience, or how tired they are.
So yes, provenance matters.
Readers, schools, publishers, and employers need fair ways to investigate credible concerns. But provenance becomes dangerous when an unusual phrase, a sudden improvement in fluency, or a detector score is treated as a verdict rather than the beginning of a conversation.
A score may justify a question.
It should not settle the answer.
The record of authorship should show judgment, not demand linguistic purity. It should show where the question began, which sources were chosen, what was rejected, where the argument changed, and who is willing to take responsibility for the result.
That seems to me a more honest account of authorship than asking whether every word was typed without assistance.
Writers have always been shaped by other people. By teachers, editors, books, conversations, countries, and languages they never entirely mastered.
The real question is not:
Did you write every word alone?
It is:
Can you show how this work became yours—and are you willing to stand behind what it became?
Perhaps that is what a voice is.
Not something that remains untouched by other languages, people, or tools, but the pattern of judgment that remains recognisably yours as you pass through them.
Sources: Commonwealth Foundation, “2026 Commonwealth Short Story Prize Update”; subsequent reporting on the authorship inquiry; Lucio La Cava et al., “Authorship Attribution in Multilingual Machine-Generated Texts,” Proceedings of ACL 2026.
First published on Substack on Aug 12, 2026 (JST). Archived here as part of the Fabio Caipirinha permanent archive (published on this site: Oct 15, 2026). Images from the original are omitted here pending rights clearance. Original: https://fabiocaipirinha.substack.com/p/a-writer-may-soon-need-to-keep-receipts
Cover image: Page from Leonardo da Vinci’s Codex Leicester, written in mirror script (Public domain, Wikimedia Commons — source)
