Authors turn to digital paper trails to fight AI authorship allegations

The rise of AI accusations

The publishing world is currently grappling with a wave of suspicion regarding the use of generative artificial intelligence. This tension reached a peak recently when the Hachette Book Group canceled a US publication run for the novel Shy Girl. The decision followed social media speculation that the book was largely produced by AI, a claim bolstered by a report in The New York Times.

The controversy centered on the work of Mia Ballard, who denied using AI and attributed any assisted changes to an editor. However, the detection tool Pangram reportedly flagged 78 percent of the novel's text as AI-generated. This incident has sparked a broader debate about how authors can defend their reputations when accusations of using LLMs (large language models) can damage a career instantly.

Novelist Andrea Bartz, who led a class-action suit against Anthropic, warns that the emergence of AI creates a climate where anyone can be accused of using it. She describes the situation as a potential witch hunt, noting that there is currently no agreed-upon standard for proving human authorship. Without a clear system, writers find themselves in a precarious position where a single accusation can lead to canceled contracts and damaged credibility.

Tools for proving provenance

In response to this growing distrust, several software companies are developing tools to help authors document their creative process. Platforms such as Workings, Blindproof, Wintertext, Human Ink, and ObraVera aim to provide a verifiable record of how a piece of writing came to be. These tools can track typing rhythms, document draft history, and even encrypt or fingerprint the final work with unique codes.

The goal of these technologies is to create a digital paper trail that demonstrates provenance. This means an author can point to a specific record of their work to prove it was created by a human, whether it was written entirely by them or with some AI assistance. These records could also serve as a defense against unauthorized scraping, where a company might use an author's work to train an AI model without permission or metadata.

However, these tools are not perfect and face significant technical hurdles. Max Spero, cofounder of Pangram, noted that a writer could still hand-type AI-generated text or use an auto-typer to bypass these systems. Furthermore, the messy reality of writing—which often involves multiple files, voice notes, and back-and-forth drafts—is difficult for a single app to capture perfectly in a linear way.

Industry reactions and hurdles

The Authors Guild, a professional organization for writers, has expressed mixed feelings about these new security nets. While the guild's director of policy and advocacy, Umair Kazi, noted that many companies have pitched these services to the organization, the CEO, Mary Rasenberger, called the requirement to preserve such receipts an 'insulting' burden. Rasenberger argues that authors should not be forced to jump through hoops to prove their own imagination and creativity.

Rasenberger also warns against overconfidence in AI detection tools, which can be fooled or produce false positives. She shared instances where human-authored work was incorrectly flagged by Pangram, citing various factors that can trick the software. She emphasizes that while the tools exist, they should be viewed as a security net rather than a mandatory standard for the literary community.

From a business perspective, many experts believe that institutionalizing these verification systems may be impractical. Jane Friedman, a publishing business expert, suggested that publishers are unlikely to adopt any verification method that costs extra money or time. Because the publishing industry is historically slow to adopt new technologies from outsiders, it may be nearly impossible to reach a shared standard for human authorship verification.

The risk of a 'witch hunt'

The social consequences of these allegations are already impacting how readers interact with new books. Andrea Bartz reported that some readers have begun review-bombing books on Goodreads simply because they felt the prose sounded like AI. Some consumers have even expressed a desire to stop reading any books published after 2022 to ensure they are not consuming AI-generated content.

This trend creates a significant risk for the entire publishing industry, as it could lead to a mass boycott of new works. Bartz expressed concern that the zeal of some readers could tank the industry by making it impossible for authors to be trusted. The fear is that the very presence of AI in the market makes it easier for people to dismiss human creativity as 'AI slop' without any real evidence.

The irony of this situation is highlighted by Bartz's own upcoming work, which satirizes the use of AI by a blocked writer. Because she intentionally wrote portions of the book to sound like an LLM—including fake transcripts and marketing materials—she faces the risk of being flagged by the very systems she is highlighting. This highlights the difficulty of navigating a landscape where the line between human-authored satire and machine-generated text is increasingly blurred.