Steven Rosenbaum revises “The Future of Truth” after AI quote errors
A line-by-line human review found 26 quotations that were corrected, paraphrased or removed from the revised book
Steven Rosenbaum’s revised edition of The Future of Truth is scheduled for publication Oct. 6 in paperback and e-book editions, according to BenBella Books, after a review found 26 quotations in the original book that could not withstand scrutiny. The new version corrects, paraphrases or removes the problematic material, following a controversy over AI-generated or misattributed quotations in a book about artificial intelligence and reality.
Rosenbaum described the process in a Fortune report, saying two human fact-checkers reviewed the manuscript line by line and traced the 26 quotations back to primary sources. The revision includes an Author’s Note, corrected citations and claims, and a new chapter, “Hallucinations and Repercussions.” It also removes every blurb and the foreword.
Among the errors was a quotation attributed to Kara Swisher that she had not said. Rosenbaum confirmed a correct quotation with Swisher for the revised edition. Questions initially arose over a passage attributed to an anonymous DARPA researcher, which New York Times reporter Ben Mullin challenged; the Times subsequently reported that the book included quotations made up by AI.
Rosenbaum said he used AI to organise research, summarise documents and refine language, and had disclosed that use in the acknowledgments. He maintained that AI did not write the book and that he remained responsible for its authorship. He said obvious errors had been caught during editing, while plausible-sounding ones survived, and accepted responsibility for failing to identify them.
The original book went through five editorial rounds and more than 4,000 tracked changes, with more than 400 citations. Rosenbaum said the review exposed a weakness in conventional editing: material that appeared credible could reach publication without being traced sufficiently to its original source.
Rosenbaum said he was conducting a full citation audit. Ars Technica reported in May that the New York Times had identified six problematic citations among 285 outside citations it examined, including three synthetic quotations with no apparent source. Rosenbaum said at the time that ChatGPT and Claude had been used heavily in research-based chapters, and that AI-derived notes were marked and passed to a fact-checker and two copy editors.
The episode had tangible consequences beyond the text itself. Rosenbaum said events were cancelled and a recorded podcast was withheld after the dispute emerged. WIRED also retracted a 1,450-word excerpt after questions about his process, though it said it had independently checked and reconfirmed the accuracy of the excerpt’s quotations and facts.
Rosenbaum began the project four years earlier as a graduate student at NYU’s Gallatin School, initially relying on Google, academic papers and books before ChatGPT and Claude were available. The revised edition adds discussion of AI fabrication in law, medicine and research, and Rosenbaum said its purpose was to repair the book and rebuild readers’ trust.