Authors Can Defend Against AI Allegations With ‘Paper Trail’ Tools. Is It Enough?

After Hachette dropped the horror novel Shy Girl this March amid claims it was largely AI-written, publishing has been living with a new kind of accusation, one that can wreck a book before anyone proves anything. Its author, Mia Ballard, denies writing it with AI and blames an editor's AI-assisted changes. WIRED reports more authors have since faced similar suspicions, including a short story prize winner, and they deny it too.

A crop of tools including Workings, Blindproof, Wintertext, Human Ink, and ObraVera now offer to record the writing process or fingerprint text as proof of authorship. The Authors Guild says writers should know about them but calls the burden insulting. The people building and critiquing these tools largely agree they are not a complete answer.

This one cuts close to home for any newsroom. The core problem is an asymmetry: an accusation costs nothing, a defense costs a lot, and there is no shared standard for proving human authorship. Bartz, who led the class action against Anthropic that settled for $1.5 billion, puts it plainly as a looming witch hunt. I think she's right that the damage lands the moment the claim does.

The tools help at the edges. A dated, hashed record of drafts gives an author something to point to, and ObraVera's founder is candid that his product proves who registered words and when, not whether AI was involved. Pangram's own CEO notes people can hand-type AI text or use an auto-typer, and that real writing happens across files, notes, and voice memos these tools never see. So the receipts are partial, and partial evidence can be spun both ways.

Detectors are the other half, and they worry me more. Pangram claims a very low false positive rate on creative writing, but the Authors Guild says it has heard of many people getting human work flagged. When a probabilistic score becomes the trigger for a cancellation or a Goodreads pile-on, the cost of every false hit falls on a person with no appeal process.

Where this leads if it scales is a surveillance tax on creative work: writers logging keystrokes to be believed, while publishers, as Jane Friedman predicts, decline to pay for any verification standard. We'd rather see publishers commit to due process before acting on AI claims, and readers treat a detector score as a question, not a verdict.

Bartz's forthcoming 2027 novel, Delete Your Account, includes passages deliberately written to sound like a chatbot, and she worries those sections could themselves be flagged. Friedman has argued the era of AI scandals may be brief as detectors improve, but that the industry is unlikely to agree on any authentication method that adds cost or time.