Steven Rosenbaum began drafting his book, *The Future of Truth: How AI Reshapes Reality*, in 2022, coinciding with the debut of ChatGPT. Although he initially refrained from using the AI, he found its capabilities improving as he continued his work into 2023 and 2024. “It never wrote a page of the book,” he clarified. “But it became a research partner, capable of providing quotes—some were impressive, others merely adequate, and a few were perplexingly inaccurate.”
Rosenbaum was diligent about verifying the information, stating, “I kept thinking I’d be really careful and double-check everything.”
However, in May, an article from the *New York Times* revealed that Rosenbaum had incorporated “more than a half-dozen misattributed or fabricated quotes” in his book, likely generated by AI. Although he had been transparent about using AI tools for research and writing, the findings were a significant embarrassment for both him and his publisher, Simon & Schuster. This incident highlighted a growing concern within the book publishing industry, which was already facing challenges posed by the rise of AI-generated texts even in the fiction market, revealing vulnerabilities in nonfiction as well.
The nonfiction publishing sector is particularly susceptible to AI-related issues because it has not traditionally prioritized fact-checking. Paul Bogaards, a seasoned public relations executive, noted that publishers are contractually unbound to verify facts before publication. They often encourage authors to independently seek fact-checking, a service that can be prohibitively expensive—ranging from $7,000 to $10,000 per book—making it unfeasible for many authors with limited budgets.
Compounding the problem is the industry's apparent lack of direction on addressing this significant risk. Literary agent Alia Hanna Habib expressed the confusion among publishing professionals, stating, “For every contract, there is a conversation, and it never really feels like anyone has the right answer.” An editor at a prominent publishing house echoed this sentiment, highlighting the chaos surrounding editorial responsibilities.
Many in the industry suspect the problem may already be widespread. A senior nonfiction editor remarked, “It feels like everyone is passing off AI work as their own without any acknowledgment.” She noted the rise in inaccuracies within manuscripts she reviewed, attributing it to the lack of publisher-supported fact-checking.
Rosenbaum himself commented that writers today heavily rely on AI for various aspects of their work, finding it both alluring and extremely beneficial. However, many authors might challenge this perspective, as there’s a growing consensus that the industry lacks clear guidelines on acceptable AI usage. Todd Shuster, a co-founder of a literary agency, pointed out an important ethical dimension: well-meaning authors might unintentionally understate their reliance on AI for research or structural ideas, leading to greater reliance on AI-generated content in their works.
Authors contractually agree that they are the sole creators of their content, which traditionally indemnifies publishers. However, Habib noted that most publishing contracts lack specific clauses addressing concerns surrounding AI. This gap leaves publishers vulnerable when faced with allegations tied to AI-generated content, which may not unequivocally align with definitions of originality.
One editor highlighted the need for a thorough discussion on the ethical boundaries of AI in publishing. Yet, the reluctance to have such dialogues persists within the industry. Another editor shared that their company provides authors with guidance against improper AI use, but these guidelines are non-binding—thus relying heavily on trust throughout the editing process.
Agents face a similar crossroads, with some suggesting that authors simply declare their non-use of AI as a condition for submission. This raises questions of legal accountability, especially when distinctions blur between permissible research assistance and improper content generation. A literary agent voiced concern about the implications of such declarations and the risks of inaccuracies that might arise from AI use.
Some agents, like Chris Parris-Lamb from the Gernert Company, have discussed the idea of specifying their disinterest in representing texts partially or wholly produced by AI. Still, enforcement mechanisms remain problematic, as the emotional weight of societal stigmas does not adequately deter misuse.
Rosenbaum’s situation demonstrated that even diligent proofreading might miss critical errors, as he noted that errors could have easily been caught by a dedicated fact-checker. After the controversy, he has hired fact-checkers for a revised version of his book. While he argued that the *Times* overemphasized the issue, he acknowledged four significant inaccuracies in a work about truth—making him a high-profile example of these AI-related challenges.
Transparency about AI use remains a contentious topic. Shuster noted that failing to disclose AI assistance could raise copyright issues, particularly since AI-generated text can resemble existing works, complicating the defense of fair use.
Though AI-detection tools have improved, their reliability remains imperfect, with concerns about misidentification of the authorship of text. “I'm concerned about false positives,” commented Parris-Lamb, though he recognized a movement toward acceptance of these tools, particularly one named Pangram that claims high accuracy. Still, the expense involved in running every manuscript through such software presents a continuing obstacle, as the industry tends to resist additional costs.
In summary, the onus of these challenges seems to rest largely on the authors and their agents, leaving a lingering uncertainty about the future of integrity in nonfiction publishing amidst the rise of AI.
