AI Doesn’t Absolve You of Getting Facts Right the First Time

This article is republished from Newsmax with the author’s permission.

In the rapidly expanding world of artificial intelligence (AI), people can be thought of as belonging to one of two groups: AI immigrants and AI natives.

To explain, I’m an AI “immigrant” because I spent much of my life before artificial intelligence became part of everyday experience.

AI “natives,” by contrast, have grown up in a world where AI is woven into daily life and have never known a time without its pervasive influence.

One difference I’ve observed is that AI immigrants like me tend to approach artificial intelligence with a healthy measure of skepticism.

Having experienced a world without AI, we are less inclined to accept its output uncritically. AI natives, on the other hand, are often more comfortable relying on AI and may accept its responses with little skepticism.

AI natives who don’t heed the warnings given by large language models do so at their own peril. OpenAI explicitly states, “ChatGPT can make mistakes. Check important info.”

During the nuclear arms negotiations of the last century, Ronald Reagan famously said, “Trust, but verify.”

With large language models, a better maxim is, “Verify, then verify.”

AI may provide a citation, but the citation itself can be inaccurate, fabricated, or point to unreliable information.

Even when a reference appears legitimate, it should be checked independently.

Those who fail to adopt this “verify, then verify” discipline risk consequences ranging from embarrassment to serious damage to their professional reputation or even their careers.

Examples are numerous, so let me highlight just a few.

One of the most prominent areas in which artificial intelligence has been misused is the scientific and scholarly literature. In academia, professors face relentless pressure to “publish or perish.”

The number of papers produced and the amount of grant funding attracted are today’s academic currency by which professors’ careers are measured.

Promotions, tenure decisions, and annual salary increases often depend more on immediate productivity than on lasting impact.

As a result, quality too often takes a back seat to quantity.

The temptation to rely uncritically on AI is strong — but the consequences can be severe.

Fabricated citations, nonexistent references, and confidently stated falsehoods can undermine the integrity of both the research and the researcher.

A particularly ironic example is the use of AI to fabricate content in a journal devoted to ethics. In 2025, a paper published in the Journal of Academic Ethics was found to contain numerous nonexistent references after a whistleblower raised concerns.

The journal investigated allegations that generative AI had been used to produce fabricated citations, highlighting the very ethical issues the publication was intended to address.

Less surprising is the use of AI to fabricate the contents of papers presented at AI conferences.

A striking example occurred at the 2025 NeurIPS conference, where researchers identified 100 fabricated citations across 53 accepted papers. Even a premier AI conference failed to detect AI-generated fabrications in the peer review process.

A large-scale 2026 analysis of the biomedical literature indexed in PubMed estimated that roughly 2,800 published papers contained fabricated references.

More troubling still, the investigators found that more than 98% of those papers had not yet been corrected, retracted, or otherwise addressed by their publishers.

The trend appears to be accelerating. STAT News noted that the number of AI-generated fraudulent citations entering academic publications increased approximately six-fold between 2023 and 2025.

Even the White House is not immune. In 2025, analysts examining the Make America Healthy Again (MAHA) report identified faulty AI-generated citations. Even high-profile government documents are vulnerable to careless use of generative AI.

The legal profession is likewise guilty of the misuse of artificial intelligence. Attorneys have repeatedly submitted unvetted AI-generated filings containing fabricated cases, fictitious quotations, and nonexistent legal citations. Here are just a few examples.

There is currently no reliable estimate of how many AI-generated fabricated legal citations have escaped detection.

Undetected cases are difficult to measure.

However, I suspect the known cases represent only a fraction of the total and that judges typically discover fabricated citations only when opposing counsel or a law clerk independently checks.

Large language models are incredible tools. I routinely use them as an alternative to a Google search, and I readily confess that I used ChatGPT to help locate many of the references cited in this article.

But I also followed my own advice: every citation herein was independently checked and verified.

When working with AI, “trust, but verify” is not enough.

The better rule is, “Verify, then verify.”