Horror novel Shy Girl lost its US publication run after Hachette Book Group acted on claims that about 78 percent of its text was AI-generated, according to detector Pangram. Author Mia Ballard denied writing the book with AI and blamed AI-assisted edits by an editor, while pursuing legal action. The case turned a publishing scandal into a business problem: an accusation alone can halt distribution and damage reputation. That risk explains demand for tools that document how a text was created.
How the Shy Girl case created a market for proof
Shy Girl had already been released in the UK when Hachette discontinued it there and canceled the upcoming US run, following social-media speculation and a Times report. Pangram assessed a large share of the text as machine-generated, although such detectors are described as robust but not infallible. Since then, other fiction writers faced similar suspicion, including a short-story prize winner and an author whose crime novel drew bids in the millions. Both denied the allegations. Novelist Andrea Bartz, lead plaintiff in the class action over Anthropic training data that ended in a $1.5 billion settlement, said the episode could trigger a witch hunt because anyone can now accuse anyone.
New platforms Workings, Blindproof, Wintertext, Human Ink and ObraVera promise a verifiable record of the writing process. Some encrypt drafts, fingerprint text or typing rhythms with unique codes, or track work inside Google Docs. ObraVera can store material with a timestamp and crypto hash, allowing later searches to show that particular words were created at a particular time by a registered user. The broader goal extends beyond authorship disputes to provenance when text is scraped into AI training sets without attribution or metadata. The record is therefore positioned as evidence of ownership as well as process.
The dispute unfolds against two larger publishing pressures: mass scraping of copyrighted books for model training and fragile trust in detection. Bartz published five thrillers since 2019, became a New York Times bestselling author, and saw work optioned by Netflix and Hulu, yet even established writers lack an agreed standard for proving human authorship. Authors Guild CEO Mary Rasenberger said detection tools can be fooled and produce false positives, despite Pangram citing a 0.01 percent false-positive rate on creative writing. Pangram cofounder Max Spero acknowledged that hand-typing AI text or using an auto-typer can evade process tracking, while drafts spread across files, emails and voice notes often escape capture.
What process tracking means for content businesses
For publishers, agencies and corporate content teams, process logs offer a practical way to reduce commercial risk around commissioned text. A timestamped draft history can support decisions on whether to proceed with publication, defend a title if readers allege AI use, and clarify responsibility between author, editor and contractor. The benefit is larger for organizations handling many manuscripts, where one canceled run carries inventory, rights and marketing costs. Smaller publishers and independent authors may use the same apps as low-cost insurance, although their counterparties may not accept every format of proof.
The limitation is that tracking shows custody and timing, not necessarily truth about AI involvement. ObraVera technical founder Brady Ridgway said the system records a user declaration that work was human-only or AI-assisted, but cannot confirm whether that declaration is accurate. Publishing expert Jane Friedman argued that agents and publishers are unlikely to adopt verification that costs extra money or time, and that agreement on a shared standard would be difficult. Buyers should therefore ask what exactly is logged, whether editing stages are covered, how manipulation is prevented, and whether a publisher or insurer will treat the record as sufficient.
The marker to watch is whether major publishers or literary agencies name an accepted form of authorship evidence during 2026-2027. If submission guidelines or contracts begin to reference specific logs, timestamps or disclosure rules, process tracking will move from optional protection to procurement requirement. If no common approach emerges, disputes will continue to be settled by detectors, public pressure and canceled runs. Bartz explores that tension directly in her forthcoming 2027 book Delete Your Account.
