TL;DR
Romy Wood, a Cardiff writer learning Welsh, entered the National Eisteddfod’s Daniel Owen Memorial Prize under a pseudonym and heard adjudicators tell a live S4C broadcast they suspected her novel had been produced by a machine and that the author had not troubled to read it back. Wood denies using AI. The Eisteddfod requires entrants to declare they have not used it but operates no system for checking, and confirmed her entry was never tested because it had not been placed in the higher category.
An accusation with no verification behind it
The three adjudicators — Meinir Wyn Edwards, Llwyd Owen, Lleucu Roberts — called it “disgraceful” that AI had been used by one of nine entries to a contest carrying this much weight. Edwards told the ceremony the book lacked any structure or arc at all, and that clumsy, flawed writing rendered it close to unreadable. The entry was placed last.
Wood, a published author writing in Welsh for the first time, described watching a minute of insults, sarcasm and laughter directed at work drawn from her own family history. She has written to the Eisteddfod asking whether that is acceptable conduct in a live broadcast, and says she has had no reply. Speaking Welsh with any confidence, she said, is something the experience has taken from her.
The adjudicators later apologised for causing distress while maintaining their unanimous view that some form of AI had produced the work. The Eisteddfod said it respects adjudicator independence, wants feedback delivered constructively, and regards encouraging learners to compete as a vital principle.
The gap between suspicion and evidence
Nothing in this process established anything. Three readers formed an impression from prose they found clumsy, and that impression was broadcast as a finding. The organisation’s own account confirms the mechanism that might have tested it — an AI check — is reserved for higher-placed entries, so the works most likely to attract a public accusation are precisely the ones never examined.
That pattern is not confined to literary prizes. A UK crime novel deal collapsed earlier this month over AI questions nobody could resolve either way, and the same dynamic runs through university misconduct panels and recruitment screening. The cost of a false positive falls entirely on the accused, who has no practical means of disproving it.
Looking forward
For UK employers and institutions writing AI policy, the operative lesson is procedural rather than technological. A declaration requirement without a verification route produces exactly this: confident public judgements resting on style, applied unevenly, with no appeal. Rebecca Thomas won the prize for her novel Ynys Gwales.