A Book About Truth Quoted Scholars Who Never Said Those Things

What happened

Steven Rosenbaum published a nonfiction book called The Future of Truth and used AI tools while writing it. When the New York Times reviewed sections of the finished book, reporters found more than half a dozen quotations attributed to named writers and scholars that were either AI-generated or wrongly attributed to the people named. A book whose stated subject is the integrity of information contained statements that its cited sources had never made.

Rosenbaum had used ChatGPT and Claude during the writing process. At some point in that workflow, synthetic text entered the manuscript as sourced quotation. The passages were not vague or anonymous. They were attributed to real, named writers and scholars whose professional authority was the reason for citing them. Readers who trusted those names encountered ideas those people had not expressed, credited to them as if they had.

The specific failure here is not that a writer made careless mistakes. Writers have misremembered quotes and garbled sources for centuries. What AI-assisted drafting changes is the mechanism and the scale. A language model produces plausible-sounding text in the style of a known thinker without any underlying source. That output is visually identical to a real quotation once it is on the page. Editors reviewing for argument and readability have no reliable signal to distinguish one from the other, and neither do readers.

Rosenbaum acknowledged the errors as accidental and said affected passages would be reviewed for correction. That response is the expected one, and it places the burden of correction on the author after the fact. The Times's reporting is what surfaced the problem. The editorial process, including whatever review the book received before publication, had not caught it. The correction came after distribution, not before.

The gap this incident exposes is verification at the point of production. No standard mechanism currently exists for a publisher, editor, or fact-checker to confirm which passages in a manuscript originated in primary sources and which were generated by a model trained to produce plausible text. Nonfiction publishing depends on the assumption that attributed quotations are real. When that assumption rests on trust rather than any traceable record, a single change in one author's workflow is enough to break it. A provable record of what a system generated, alongside documentation of what a human verified against a source, would shift that standard from assumption to evidence, and would surface problems before a book reaches readers rather than after.

Reported impact

Affected parties

Not publicly disclosed

Harm type

Not publicly disclosed

Scale

Not publicly disclosed

Financial impact

Not publicly disclosed

Regulatory action

Not publicly disclosed

Classification

Organization

Not publicly disclosed

AI system

Not publicly disclosed

Industry

Not publicly disclosed

Country

Not publicly disclosed

Provider

Not publicly disclosed

Incident type

Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

• No controls mappedNot publicly disclosed

Control mapping is analytical. It does not state that any control would have prevented the incident.

Sources and evidence

This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.