Glonce

Paper-trail tools promise authors proof of human writing after Shy Girl

toolsThe Glonce editorsWired🔥51Glonce rating51Impact40Novelty45Relevance80Surprise35Players25Credibility70
Paper-trail tools promise authors proof of human writing after Shy Girl

WIRED opens with novelist Andrea Bartz: five thrillers since 2019, a New York Times bestseller, work optioned by Netflix and Hulu, and lead plaintiff in the class action that led Anthropic to agree, last year, to pay $1.5 billion in what the article calls the largest copyright settlement in US history. Her worry now is a different threat: authors being accused of using AI themselves.

That threat became real this past March. After widespread social media speculation that the horror novel Shy Girl was largely AI-generated, and a Times article about the claims, Hachette Book Group canceled an upcoming US run and discontinued the book in the UK, where it had already been released. Pangram, which WIRED calls a robust but by no means infallible detection tool, rated some 78 percent of the text as AI-generated. Author Mia Ballard denies using AI. She says an editor was responsible for AI-assisted changes to a previously self-published version, and that she is pursuing legal action. Since then, several more fiction authors have faced AI suspicions, one after winning a short story prize and another after publishers bid into the millions for his crime novel. Both deny it.

Bartz says that once accusations start, the damage is done. She felt the scandal would lead to "a potential witch hunt" in which anyone can accuse anyone, and authors have little recourse: "There's no agreed upon standard" for proving human authorship.

A group of new software tools hopes to fill that gap. Workings, Blindproof, Wintertext, Human Ink and ObraVera promise to track and document an author's writing process, in some cases encrypting and fingerprinting the text or typing rhythms with unique codes, to produce a record of authorship, whether entirely human or AI-assisted. The idea is also to show provenance when text is scraped into AI training sets without attribution.

Mary Rasenberger, CEO of the Authors Guild, says the guild cannot recommend one tool over another but wants writers to know they exist as something of a security net. Many of the companies have approached the guild to pitch their services, per Umair Kazi, its director of policy and advocacy. Rasenberger also calls the need to jump through this hoop insulting and a burden, and says the guild does not feel it should be required for authors to preserve receipts. She argues AI-generated submissions are rarer in publishing than people think, and warns of a climate of distrust and overconfidence in detectors. Pangram claims an industry-leading false positive rate of 0.01 percent on creative writing, but she says the guild has heard of many cases where people ran their own human-written work through Pangram and were told it was AI-generated.

Process trackers have limits too. Max Spero, cofounder and CEO of Pangram, says his company offers a product that tracks writing in Google Docs and many users find it helpful, but paper-trail tools are not a perfect catch-all: someone could hand-type AI text or use an auto-typer. Real writers also work across multiple files, swap drafts and dictate voice notes, and not all of that is captured. Brady Ridgway, technical founder of ObraVera, says his platform is built more for copyright: with a time stamp and crypto hash it can show that certain words were created at a given time by a given registered person. But it is up to that person to declare whether they used AI, and the software cannot confirm the declaration is true. He points to the May dispute over a writer accused of winning a Commonwealth Short Story Prize with an AI-generated submission: software like his would only show who submitted the story and when.

Jane Friedman, an author and publishing business expert, argued in her newsletter The Bottom Line that the era of AI scandals is likely to be short-lived, given improvements in Pangram and similar detectors, and that authors have seen what can happen if they pass off AI writing as their own. She tells WIRED that agents and publishers are unlikely to adopt any verification method that costs extra money or time, and that agreeing on a shared standard would be "like herding cats" because publishing is slow to adopt technology, especially from industry outsiders. Integration into production workflows and faked or manipulated certification are further hurdles. Authors may pay for peace of mind, she says, but cannot trust that agents and publishers will accept proof from any given app.

With no clear solution, writers are left to fend for themselves. Bartz says she has heard of people review-bombed on Goodreads because a stranger thought a book sounded like AI, and of readers vowing to read nothing published after 2022. Her own forthcoming 2027 book, Delete Your Account, tackles generative AI: a blocked writer prompts a chatbot to imitate her favorite thriller and ends up trapped in that story. Bartz wrote parts meant to sound like LLMs, including made-up transcripts and AI-sounding marketing materials, and wonders what she will do if any of it is flagged as AI-written.

Key facts

  • Hachette Book Group canceled the US run of Shy Girl and discontinued it in the UK in March after AI allegations; Pangram's tool rated some 78 percent of the text as AI-generated, and author Mia Ballard denies using AI.
  • Workings, Blindproof, Wintertext, Human Ink and ObraVera promise to document an author's writing process, in some cases by encrypting and fingerprinting the text or typing rhythms.
  • The Authors Guild will not recommend any one tool and says authors should not be required to keep such receipts; it also says human work has been wrongly flagged as AI by Pangram, which claims a 0.01 percent false positive rate on creative writing.
  • Pangram's Max Spero says tools can be fooled by hand-typed AI text or auto-typers; ObraVera's Brady Ridgway says its records show who wrote what and when, but the author's AI declaration cannot be verified.
  • Jane Friedman expects agents and publishers to reject any verification that costs extra money or time, and thinks the era of AI scandals may be short-lived.

Why it matters

Authors have spent years worried about AI companies training on their books. The Shy Girl case shows a second front: an accusation alone, driven by social media and detector scores, can end a book's publication. Bartz's point is that once the charge is made, the damage is done, and there is no agreed standard for proving human authorship. The paper-trail tools are the market's first answer, and the article asks whether they can work.

Who it affects

Fiction authors are the most exposed, including those who are innocent: the article cites several writers accused after a prize win or a big book deal, and Bartz fears being flagged for her own satirical, LLM-style passages. Publishers, agents and readers are also drawn in, as seen in Goodreads review-bombing and readers who say they will avoid anything published after 2022. The tool makers, the Authors Guild and detector companies like Pangram are the other parties.

How to use it

The Authors Guild cannot recommend any one tool but wants writers to know they exist as a security net. The tools record the writing process, and some fingerprint the work or typing rhythms; Pangram also offers a product that tracks writing in Google Docs. ObraVera is aimed at tracing reuse of material stored in its database by a registered user, with a time stamp and crypto hash. The article gives no price, user numbers or adoption figures for any of these tools.

How solid is it

This is WIRED's reported feature, with named sources: the Authors Guild's CEO and policy director, Pangram's CEO, ObraVera's technical founder and Friedman. Several key numbers are claims by interested parties. The 78 percent figure is Pangram's tool output, not a verified fact about the book, and the 0.01 percent false positive rate is Pangram's own claim. The article does not say whether any publisher or agent accepts output from these tools as proof; Friedman only predicts they are unlikely to.

Risks and caveats

Process trackers can be gamed: Spero says someone could hand-type AI text or use an auto-typer, and the article describes an ongoing cycle of vulnerabilities and fixes on both sides. They also miss real workflows such as multiple files, shared drafts and voice notes. ObraVera cannot check whether a user's declaration of no AI use is true. Friedman warns that authors cannot count on publishers accepting any app's proof, and that faked or manipulated certification is a further problem. The Authors Guild adds that detectors themselves produce false positives. The outcome of Ballard's announced legal action is not given.

“If any of this costs extra money or extra time, forget it”

— Jane Friedman, author and publishing business expert, to WIRED