“`html
Andrea Bartz is a prolific novelist with an impressive track record. Since 2019, she has released five suspense novels, achieved bestseller status on the New York Times list, sold the rights to her works to platforms like Netflix and Hulu, and received endorsement from the influential Reese Witherspoon book club.
In addition to her literary accomplishments, Bartz has a legal reason to be in the spotlight; she was the lead plaintiff in a class action lawsuit against Anthropic. The case accused the company of downloading millions of copyright-protected books from piracy sites to use as training data for their models. Last year, Anthropic agreed to a historic $1.5 billion settlement, marking the largest copyright agreement in U.S. history.
While Bartz and many other writers remain vigilant against the exploitation of their work, a new threat related to artificial intelligence (AI) has emerged.
Writers Accused of AI Usage
What happens when authors themselves are accused of resorting to AI? The recent case of Peruvian author David Silva Rojas illustrates this troubling trend. Rojas won the prestigious International Short Novel Prize Ventosa-Arrufat for his work, “Donde el agua trabaja,” which competed against 2,204 manuscripts from 22 countries. However, on October 23, just before its debut in Spain, an investigation by El País revealed that Pangram, a robust AI detection tool, attributed 99% of the text to AI.
In March, the horror novel “Shy Girl” also came under scrutiny. Following a report in The Times that raised concerns about potential AI involvement in its writing, Hachette Book Group cancelled its upcoming U.S. release and withdrew it from the market in the UK, where it had already been published. Pangram indicated that approximately 78% of the text may have been generated by AI.
Mia Ballard, the author of “Shy Girl,” has denied allegations of using AI to write the novel, claiming that an editor was responsible for changes made with the aid of AI in a previous version she had published. She has also stated her intention to pursue legal action. Since the controversy erupted, several other fiction writers have faced AI-related suspicions, with one author being questioned after winning a short story award and another after publishers made multimillion-pound bids for their noir novel. Both have staunchly denied these accusations.
The Impact of Accusations on Authors
Bartz indicates that once allegations arise, the damage is often already done. “When the Shy Girl scandal broke, I immediately felt that this could lead to a potential witch hunt,” she explains. The inevitable incursion of AI into the publishing world, she elaborates, “will have the unintended effect of allowing anyone to accuse anyone else of using AI,” and currently, authors possess limited means of defending themselves against such claims.
“There is no consensus standard for proving human authorship,” Bartz points out.
A number of new software tools are being developed with the aim of changing this scenario. Platforms such as Workings, Blindproof, Wintertext, Human Ink, and ObraVera promise to provide means for tracking and documenting an author’s writing process, in some cases encoding and assigning a “digital fingerprint” to written work or writing patterns through unique codes to create an authorship record (whether entirely human or AI-assisted). The intention is not only to create a documentary record that writers can refer to if questioned about their use of a large language model (LLM) but also to establish provenance when, for instance, text is extracted for inclusion in AI training datasets without attribution or identifying metadata.
Emerging Tools for Authorship Verification
Mary Rasenberger, CEO of Authors Guild, a professional organisation focused on free speech and copyright protection, states that while the association cannot endorse any of these tools over others, it wants writers to be aware of their existence as a safety net. Many of the companies have approached Authors Guild to present their services, according to Umair Kazi, the association’s director of policy and advocacy.
Simultaneously, Rasenberger explains that it is a “burden” to have to navigate this process: “We don’t believe it should be mandatory for authors to keep receipts.” She adds, “If what you are selling is your words, your voice, your imagination, and your creativity, it makes no sense to use AI,” noting that authors take pride in their own work, making low-quality AI-generated outputs much less common in the publishing world than one might think.
“I’ve spoken with editors, particularly in literary non-fiction and fiction, and some say, ‘I never see AI-generated manuscripts; if I did, I would recognise it,'” Rasenberger argues.
A Climate of Distrust Among Authors
Despite Pangram claiming an industry-leading false positive rate of 0.01% when used for creative writing, Rasenberger notes that the Authors Guild has been made aware of numerous instances where individuals submitted their own human-written works to Pangram and received results indicating they had been generated by AI. Various factors can lead an AI detector to conclude that something was AI-generated when it was not.
Similarly, a process-tracking application is not infallible; the AI arms race means there is a continuous cycle of vulnerabilities and fixes on both sides. “We offer a product that tracks writing in Google Docs, and many users find it helpful. That said, change-tracking tools are not a perfect solution for everything. For example, someone could handwrite text generated by AI or use an auto-writing program,” states Max Spero, co-founder and CEO of Pangram. Spero highlights that the inherent chaos of the writing process presents additional challenges. “The average writer doesn’t just sit down and write a book. They work with multiple files, swap drafts, and use voice notes while on the go. None of this is captured by writing process tracking tools.”
Protecting Copyright Versus Human Authorship
These tools may be more focused on safeguarding copyright than on determining whether content was created by a person or by AI. Brady Ridgway, technical founder of ObraVera, explains that his platform can create cryptographic “fingerprints” from patterns found in both text and voice recordings. The goal is to track the reuse of material that a user has previously registered in their database.
Using a timestamp and a cryptographic hash, the platform can verify when and who registered a particular piece of content. Thus, Ridgway explains, if that material “later appears somewhere it shouldn’t be or someone else claims it’s theirs, we can search the database and say: ‘These words were created at this moment by this person.'” This can demonstrate provenance and ownership, Ridgway asserts, although it remains up to the registered individual to declare whether they used AI in creating the text. He refers to the uproar that emerged in May surrounding a writer accused of winning the Commonwealth Short Story Prize with an AI-generated work. “Everyone seems to agree that he is the person who submitted the story. That’s all the software would do. It would state that someone, on that date, declared it to be an exclusively human work or not AI-assisted,” Ridgway mentions. He cannot confirm whether that claim is accurate or not.
The Challenges of Institutionalising Authorship Verification
Some argue that attempting to institutionalise systems for verifying human authorship is simply impractical and that it may soon become irrelevant. Jane Friedman, an author and publishing expert, recently argued in her industry newsletter, The Bottom Line, that the era of AI-related scandals is likely to be short-lived, given the advancements of Pangram and other similar detectors. She also noted that authors have already seen the consequences of trying to pass off AI-generated text as their own.
Friedman explains to WIRED that it is unlikely agents and publishers will implement any method of verifying human authorship that “entails additional costs or requires more time,” partly because it would be impossible to get them to commit to adopting a common standard. She predicts that “achieving consensus on this will be like herding cats, as the publishing industry is slow to embrace new technologies, particularly those developed by outsiders.”
The Difficulty of Detecting AI in Literature
Even if various publishers with differing needs and resources agreed on some theoretical form of authentication, logistical hurdles would still exist. Friedman points to the logistical challenges of integrating this criterion into existing production models, not to mention the ongoing issue of “falsified or manipulated certifications.” “If any of this entails additional costs or requires more time, forget it,” she concludes. And while authors may be willing to pay for peace of mind, Friedman asserts that they cannot rely on agents and publishers to accept authorship proof from any application.
In the absence of a clear solution, writers must fend for themselves. Bartz expresses concern that readers may feel deceived, believing that what they read could have been produced by AI. “I’ve heard of people receiving a flood of negative reviews on Goodreads because some stranger thought the book sounded AI-generated, and everyone else piled on the critique. I’ve seen readers say, ‘I’m not going to read anything published after 2022 because that way I can be sure it doesn’t contain generative AI.’
Interestingly, Bartz’s next book, set to be released in 2027, directly addresses AI. Its protagonist is