A newly published book about protecting truth in the age of artificial intelligence contains quotations that were misattributed or fabricated by AI tools, The New York Times reported on May 19, 2026, and the author has acknowledged the errors.
The book, “The Future of Truth” by American author and media entrepreneur Steven Rosenbaum, argues for safeguarding trust and factual accuracy as generative AI spreads. According to the Times, a check of its citations found several quotes attributed to people who never said them, along with others that were real but credited to the wrong source.
What the author says
Rosenbaum has confirmed that the book contains misattributed or artificially generated quotations and said he has begun an internal review of the text. He said he used AI tools, including ChatGPT and Claude, during research, writing and editing, but added that this did not excuse the mistakes. He has said the affected passages will be corrected in future editions.
Who was misquoted
One fabricated passage was attributed to the technology journalist Kara Swisher, who told the Times she had never said it and suggested an AI system had produced it. Several quotes attributed to the psychologist Lisa Feldman Barrett also turned out to be wrong. Barrett said the words did not appear in her book and did not reflect how she talks about the subject.
Other citations were not entirely invented. In those cases the underlying statement existed but was linked to the wrong book or the wrong author, a pattern consistent with how large language models blend and misremember sources.
A familiar failure mode
The episode is the latest in a series of cases in which professionals relied on AI-generated references without checking them, including lawyers who filed court briefs citing nonexistent cases and newspapers that published AI-written summer reading lists featuring books that did not exist. Generative models can produce plausible, confidently worded material that has no basis in fact, and when writers or institutions pass it along unverified, responsibility for the error does not rest with the software alone.
That it happened in a book whose central theme is truth and trust in the AI era has drawn particular attention in the publishing industry, where editors and publishers are still working out how to disclose and check AI-assisted writing. The practical lesson, as several commentators noted, is unchanged: AI can help with research, but verification remains a human job.
