待翻译：Book about AI and truth shipped with fake AI-generated quotes

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

Book about AI and truth shipped with fake AI-generated quotes May 2026 In May 2026, Steven Rosenbaum's The Future of Truth became the wrong kind of case study when The New York Times, The Daily Beast, The Atlantic, and Ars Technica reported that the book contained multiple fake or misattributed quotes. Rosenbaum acknowledged using ChatGPT and Claude during research, writing, and editing, and accepted responsibility for what he called improperly attributed or synthetic quotes. Reporters found a fabricated quote attributed to Kara Swisher, misattributed material connected to Lisa Feldman Barrett, and a Meredith Broussard quote placed in the wrong source. Ars reported that six outside citations had been identified as problematic. A book warning about synthetic truth managed to demonstrate the footgun in hardcover. Incident Details Severity:Facepalm Company:Steven Rosenbaum / The Future of Truth Perpetrator:Author Incident Date:May 2026 Blast Radius:A nonfiction book about AI and truth launched with multiple fake or misattributed quotes; named journalists and scholars publicly disputed attributions; future editions require correction; the author's credibility and the book's central argument were damaged. Tech Stack ChatGPTClaudeAI-assisted researchBook publishing workflow References The Daily Beast: Author Steven Rosenbaum busted using fake AI-generated quotes in book critiquing AI ↗The Atlantic: AI-writing scandals are getting very confusing ↗Ars Technica: AI put synthetic quotes in his book, but this author wants to keep using it ↗ A book about truth met the machine Steven Rosenbaum's The Future of Truth was supposed to examine how artificial intelligence reshapes reality, public trust, and media. It landed in May 2026 with exactly the kind of problem its subject matter should have made impossible to miss: fake and misattributed quotes apparently tied to AI-assisted research. The New York Times first reported the quote problems on May 19. The Daily Beast summarized the findings the same day, and The Atlantic followed with its own interview and analysis. Ars Technica also covered the case, including Rosenbaum's explanation of how the errors entered the manuscript. Rosenbaum acknowledged that the book contained a handful of improperly attributed or synthetic quotes and said he had used ChatGPT and Claude during the research, writing, and editing process. He said the use of AI did not excuse the mistakes and that future editions would be corrected after review. That is the responsible public posture. It is also a brutal publication-week problem for a book about truth. If your nonfiction book warns readers about AI making reality slippery, the footnotes need to be boringly, aggressively verified. Otherwise the book becomes a practical demonstration instead of an argument. What went wrong The Daily Beast reported three categories of attribution failure. First, the book included a quote attributed to technology journalist Kara Swisher that she denied saying. The quote itself was very much in the polished-chatbot style: grand metaphor, high-confidence abstraction, no human fingerprints except perhaps the part where the named human objected to having fingerprints assigned to it. The Daily Beast reported that Swisher told The New York Times she had never said it. Second, a section involving neuroscientist Lisa Feldman Barrett included multiple quotes tied to her book How Emotions Are Made. Barrett told the Times the quotes were not in the book and were also substantively wrong. That second part matters. A misattributed quote can be a sourcing error. A misattributed quote that also misrepresents the expert's actual view is a content error wearing a citation badge. Third, the book cited author Meredith Broussard and her book Artificial Unintelligence for a line that was apparently real but came from a different source: a 2023 Marketplace Tech interview rather than the book named in the citation. That is a less flashy error than a wholly fabricated quote, but it points to the same underlying workflow failure. The attribution chain was not checked back to the primary source. The Atlantic reported that Rosenbaum had relied on AI tools as both a resource and a conversation partner while working on the book. It also reported that he later blamed ChatGPT for the errors while still saying he could not imagine giving the tool up. That combination is the most interesting part of the incident. The author is not anti-AI, and the scandal is not that AI touched the process. The scandal is that AI was used in a context where it is famously unreliable, then its output was not verified before publication. Why quote errors are not minor A quote in nonfiction is not a vibe marker. It is evidence. When a book attributes words to Kara Swisher, Lisa Feldman Barrett, or Meredith Broussard, readers are being told those people said those things in those places. The quote anchors an argument to outside authority. If the words were invented or assigned to the wrong source, the argument is borrowing credibility it has not earned. AI systems are especially dangerous here because they are good at producing quote-shaped language. They can generate a sentence that sounds like a media critic, a neuroscientist, or a technology scholar. They can pair it with a plausible book title or interview. They can create the feeling of research without the actual research underneath it. The output looks citation-ready because the model has learned the surface form of citation-ready prose. That is the footgun. A sloppy researcher may at least know when they are guessing. A chatbot presents the guess in the house style of confidence. If the author or editor treats that confidence as evidence, the fake quote travels straight into the manuscript. The publishing process should have caught this. Quotations in nonfiction normally get checked against source material. Editors, copy editors, fact-checkers, and authors all share responsibility for making sure quoted words are real and properly attributed. The Future of Truth failure suggests that either the checking did not happen, or it happened against AI-produced notes rather than original sources. Both routes end at the same ugly place. The irony is obvious; the workflow problem is more useful Yes, a book about truth in the age of AI containing AI-generated fake quotes is almost too on the nose. The jokes write themselves, which is convenient because the book apparently had enough trouble with machine-written material already. But stopping at irony makes the lesson smaller than it should be. This incident is a clean example of how AI research assistance can contaminate serious publishing without anyone intending to deceive readers. Rosenbaum did not need to ask a chatbot to fabricate quotes. He needed only to ask it for help with research, accept material that looked plausible, and fail to chase the quotations back to original sources. That is how most professional AI failures happen. Not through cartoon malice. Through speed, trust, deadline pressure, and one missing verification step. AI makes the draft feel more complete than it is. The human process relaxes. The product ships with synthetic facts embedded in it. The correction process will probably repair future editions. It will not repair the first impression. The book's credibility now carries a permanent asterisk because readers know the author warned about synthetic reality while relying on a synthetic research layer that inserted falsehoods into the text. What should have happened The safe workflow is plain. Every quote gets traced to a primary source. If the quote came from a book, open the book. If it came from an interview, open the transcript or recording. If an AI tool supplies a quotation but cannot produce a verifiable source location, the quote is unusable. If it supplies a source location, verify that location manually. AI can be a brainstorming partner. It can help assemble a list of places to search. It can summarize notes if the notes are already verified. It cannot be the authority for who said what. That job belongs to source documents. The Future of Truth now belongs in the graveyard because it is a tidy demonstration of a larger publishing risk: when AI is allowed to stand between the writer and the evidence, the machine can invent the evidence. A book about truth should have known better. So should every publisher watching this happen in public.