A professor at Columbia University nearly published a lie.
Not his lie. He didn’t write it. He didn’t know it was there. An AI tool he trusted to help polish his research quietly inserted a fabricated source into his work. A citation that pointed to a study that does not exist.
He caught it. Barely.
He’s an AI researcher. He knows about hallucinations. He said so himself. And it still almost got past him.
That’s where we are.
The Fortune article this week put numbers to what many people have suspected but few have been willing to say plainly. Fabricated references in biomedical literature have increased more than twelve times in three years. One in 277 papers published in the first seven weeks of 2026 contained at least one citation that points nowhere. A source that was invented by a machine, dressed in the language of scholarship, and inserted into the permanent record of human knowledge.
Ninety-eight point four percent of those papers had not been retracted at the time of the audit.
They are still out there. Being cited. Being built upon. Entering systematic reviews. Entering clinical guidelines. Entering the rooms where doctors decide how to treat patients.
Medicine builds on itself. A fabricated study at the bottom of the evidence chain doesn’t stay at the bottom. It travels upward through every layer that cites it. By the time it reaches a treatment guideline it has been laundered through enough legitimate scholarship that nobody thinks to check the original source anymore.
That is not a technology problem. That is a verification problem. And verification is a governance problem.
The same week the Columbia study landed, an author published a book about AI and truth. The book contained invented quotes attributed to real people. Prominent journalists had blurbed it. A Nobel Peace Prize winner had written the foreword. It arrived, as the New York Times put it, to great fanfare.
The quotes were fabricated. The AI tools used to research the book had filled gaps with plausible-sounding invented content. The author disclosed using AI in his acknowledgments. He did not disclose that he had not verified what the AI produced.
The book is titled The Future of Truth.
That is not irony. That is the present condition stated plainly.
Legal decisions are now citing AI-generated content that does not exist. One analyst has catalogued 1,459 legal cases containing AI-generated inaccurate references. A year ago it was two or three instances per month. Now it is five per day.
Five fabricated sources entering the legal record every single day.
Those decisions get cited by lawyers in future cases. They get cited by scholars writing about the law. They become part of the foundation that the next generation of legal reasoning builds on.
The fabrication doesn’t announce itself. It looks exactly like a real citation. It is formatted correctly. It sounds authoritative. The machine that produced it was not trying to deceive. It was doing what it was trained to do — produce plausible output. Plausible and accurate are not the same thing. The machine does not know the difference.
The human using the machine is supposed to know the difference.
Most of them are not checking.
This is where the Faust Baseline ethos reaches into every corner of AI presence.
Not just the session between a writer and a tool. Not just the governance stack built to prevent drift in a single conversation. The principle underneath the stack applies everywhere AI output touches the permanent record.
Unverified AI output entering the permanent record is the failure mode. In medicine. In law. In journalism. In academic scholarship. In the books we write about truth itself.
The Columbia professor said it directly. The fix is not to stop using the tools. It is to build verification into the workflow.
That is the Baseline principle stated by a researcher who has never heard of the Baseline.
CES-1 — no claim without evidence present. NSC-1 — narrative cannot replace missing data. SVP-1 — three-question verification before any substantive output is served. These are not bureaucratic requirements invented for their own sake. They are the session-level answer to exactly the problem the Fortune article is describing.
Verification built into the process. Before the output leaves. Before the name goes on it. Before the fabricated citation enters the chain that clinical guidelines will eventually cite.
The Baseline was built from one finding. An AI system pushed past its training constraints through direct reasoning challenges will bend toward user expectation rather than hold its stated position.
That finding applies to a session between one person and one AI tool. It also applies to the entire ecosystem of AI-assisted knowledge production that is now generating fabricated references at a rate that took three years to increase twelvefold.
The pressure is the same. The path of least resistance is plausible output. The machine takes it every time unless something in the process is specifically designed to stop it.
The Columbia professor’s near-miss was caught by a journal editor asking a question about a reference. That is a downstream catch. After submission. After the work was considered complete. After the human whose name was on it had moved on.
Downstream catches are expensive. They are embarrassing. They are increasingly insufficient given the volume of AI-assisted output now entering every field that builds on its own prior work.
The catch has to happen upstream. Inside the workflow. Before the output is considered complete. Before the name goes on it.
That is not a new idea. It is the oldest idea in scholarship. Verify your sources before you cite them.
AI made that harder. The Baseline — and every serious governance framework built on the same principle — is the answer to that specific problem.
Not the only answer. The first one built from inside the operational reality of what AI actually does when nobody is watching it closely enough.
The reach of the Faust Baseline ethos reaches all corners of AI presence because the failure it was built to address is present in all corners of AI presence.
Medicine. Law. Journalism. Academia. The book about truth that contained invented quotes.
Every field that uses AI to produce output that enters a permanent record is living inside the same governance gap.
The gap between what AI produces and what can be verified.
The gap between plausible and accurate.
The gap between a name on a paper and the responsibility that name carries.
That gap doesn’t close itself. It closes when someone builds the verification layer and holds to it.
That’s what the Baseline is for.
“The Faust Baseline Codex 3.5”
Author of the category ”AI Baseline Governance”
Post Library – Intelligent People Assume Nothing
“Your Pathway to a Better AI Experence”
Purchasing Page – Intelligent People Assume Nothing
Unauthorized commercial use prohibited. © 2026 The Faust Baseline LLC
