Authors accused of using AI turn to tools that record how they write

A group of software tools now promises to give authors a record of how they wrote their manuscripts. Workings, Blindproof, Wintertext, Human Ink and ObraVera track and document the writing process, according to Wired. Some of them encrypt and fingerprint the text, or the writer's typing rhythms, with unique codes.

The tools are a response to suspicion. In publishing, a public accusation of AI use can now end a book contract. Authors usually cannot answer it, because there is no accepted way to prove that a human wrote a text. Writers are starting to collect evidence in advance, and the Authors Guild says they should not have to.

The cases behind the suspicion

In March, Hachette Book Group cancelled the planned US publication of Shy Girl, a horror novel by Mia Ballard, and discontinued it in the UK, where it was already on sale. The decision followed widespread speculation on social media and a Times article reporting the claims. Pangram, an AI detection tool, rated about 78 percent of the text as AI-generated. Wired describes the tool as robust but not infallible. Ballard denies writing the book with AI. She says an editor made AI-assisted changes to an earlier, self-published version, and she says she is pursuing legal action.

More authors have faced similar suspicions since then. One was accused after winning a short story prize. Wired names it later as the Commonwealth Short Story Prize, in May. Another was accused after publishers bid into the millions for his crime novel. Both deny the allegations.

Novelist Andrea Bartz was the lead plaintiff in the class action against Anthropic over pirated books used as training data. Anthropic agreed last year to pay $1.5 billion to settle that case. Bartz told Wired that the Shy Girl affair felt like the start of "a potential witch hunt" and that "there's no agreed upon standard" for proving human authorship. She has heard of authors being review-bombed on Goodreads because a stranger thought a book sounded like AI. Her own 2027 novel, Delete Your Account, includes passages written on purpose to sound like a language model. She does not know what she would do if those passages were flagged.

What the tools record, and what they cannot

The tools are meant to produce a record that an author can show when their work is questioned. They are also meant to establish provenance, for example when text is scraped into AI training sets without attribution.

The people who build these tools admit their limits. Max Spero, cofounder and CEO of Pangram, says his company offers a product that tracks writing in Google Docs. He also says "paper trail" tools are not a perfect catch-all: "someone could hand-type AI text, or use an auto-typer." He adds that real writing is spread across many files, emailed drafts and voice notes, and most of that is never captured.

Brady Ridgway, technical founder of ObraVera, says his platform is designed mainly for tracing reuse. It stores a time stamp and a cryptographic hash for work registered by a human user. If the text later appears somewhere else, the database can show that "these words were created at this time, by this person." Whether AI was used is left to the user's own declaration. The software records that the declaration was made. It cannot check whether it is true.

Detection has the opposite problem. Pangram reports a false positive rate of 0.01 percent on creative writing. Mary Rasenberger, CEO of the Authors Guild, says the guild has "heard of lots of instances" in which writers ran their own work through Pangram and were told it was AI-generated.

A burden placed on the writer

Rasenberger says the guild cannot recommend one tool over another. Many of the companies have pitched their services to the guild, according to its director of policy and advocacy, Umair Kazi. She wants writers to know the tools exist as a safety net. She also calls the requirement "insulting" and a "burden", and says "we don't feel it should be required" for authors to keep receipts. She asks authors not to accuse other authors, because it does not support a healthy literary community.

Jane Friedman, an author and publishing business analyst, expects the scandals to fade as detectors improve and as authors see what happens to people caught passing off AI text as their own. She doubts agents and publishers will adopt any verification method that "costs extra money or extra time". She also doubts the industry would agree on a shared standard, and she points to the risk of "faked or manipulated" certificates. Authors may pay for a tool, Friedman says, but they cannot count on a given publisher accepting what it produces.

What it means for work

For writers, proving authorship is turning into a new administrative task: running tracking software, keeping drafts, registering hashes. It adds work for the people who make the product and saves nothing for the people who might doubt it. These records can show when a text was created and who registered it. They cannot show that no model was involved, because a typed copy of machine text leaves the same record as original writing. Detectors that judge the finished text and tools that log the process each have a known way to fail. There is still no standard that publishers have agreed to accept. Until there is one, an author's defence against an accusation depends on evidence that nobody is obliged to recognise.