AI & Institutions

The Circularity Trap: Credential Theater in the Age of AI

When the applicant’s machine writes to the university’s machine, and the student’s machine writes to the teacher’s machine, the paperwork multiplies and the truth quietly disappears.

I have spent a career reading financial statements built to survive an audit rather than to describe a business. So when I look at what is happening in admissions offices and classrooms right now, I recognize the pattern immediately. It is not new. Only the paper has changed.

The original version of this problem showed up first in the workplace. An employee, asked to write a self-assessment, runs it through an AI model to sound more strategic, more metrics-driven, more promotable. A manager, asked to review it, runs the response through another AI model to generate feedback that sounds thoughtful and specific. Two machines produce a compatible narrative. A human signs it. Nobody involved has actually evaluated anything. The organization ends up with more paper and less truth. That is not progress. It is theater with better grammar.

The Admissions Version

Now watch the same mechanism move into college admissions. An applicant, coached by AI, produces a personal statement engineered to hit the emotional and structural beats that admissions software and overworked readers reward — the adversity arc, the resolved conflict, the tidy epiphany. On the other side, universities under application-volume pressure increasingly lean on AI-assisted first-pass review, algorithmic scoring, or simple pattern-matching to triage tens of thousands of essays into a manageable pile.

What you get is a machine writing to satisfy a machine, with a human admissions officer skimming the compatible output somewhere in the middle and calling it holistic review. The applicant never had to produce an original thought. The university never had to genuinely assess one. Both sides can point to a completed, defensible-looking process. Neither side has learned anything true about the applicant sitting behind the essay.

The Classroom Version

The same loop is closing inside the classroom itself. A student assigns the homework, the essay, the lab report to an AI model. A teacher, buried under a caseload that no accreditation body has adjusted for the new reality, runs the submission through an AI detector, or increasingly, through an AI-assisted grading tool that reads for structure, citation, and coherence rather than for understanding. Detector software trained to catch machine writing is checked against generator software trained to evade detection — an arms race between two models, with the student’s actual command of the material as a bystander to the whole exchange.

The grade gets recorded. The transcript gets built. The GPA compounds into a scholarship, a class rank, a resume line. At no point in that chain did a human being confirm that a mind absorbed anything. The credential is real. What it certifies is not.

Call It What a Restructuring Man Would Call It

In my old world, there is a name for a transaction that generates paper volume without generating economic substance: round-tripping. Two related parties trade an asset back and forth, booking revenue on both ends, until the balance sheet looks active and growing while nothing of value has actually changed hands. Regulators treat it as fraud when the parties are corporations. Nobody has yet named it when the parties are a chatbot writing an essay and a chatbot grading one, but the mechanics are identical. Volume is being manufactured. Substance is not.

Every credentialing system runs on a promise: this diploma, this GPA, this admission letter is a reliable signal that something real occurred behind it. AI-to-AI circularity doesn’t just weaken that signal — it counterfeits it at scale, on both the input and the output side of the ledger, with a human notary in the middle who never checked the collateral.

What Actually Restores the Signal

The fix is not banning the tools. That fight is already lost and, frankly, not worth winning — AI-assisted drafting and AI-assisted feedback both have legitimate uses. The fix is putting a verified, un-outsourceable human checkpoint back into the process: proctored writing samples, oral defenses of submitted work, in-person problem-solving, interviews where an applicant has to think in real time rather than submit a polished artifact. Anything that cannot be generated by one model and cleared by another restores the thing the whole system is supposed to be measuring in the first place.