The Parrot Wore a Suit

The copy shop had a machine in the corner that sounded like it was trying to cough up a lung.

A young man in a blue shirt kept feeding it pages from a folder marked IMPORTANT in red ink. The machine swallowed them, flashed a little green light, and spat out copies with a black streak down the left side. Every page came out wounded in the same place.

He looked at the streak. He looked at the machine. Then he stacked the bad copies neatly and kept going.

That is civilization, more or less.

The defect is visible. The machine is making the same mistake every time. The human being sees it happen, understands it is happening, and continues producing official-looking garbage because the meeting starts in twenty minutes and nobody wants to be the person who stops the procession.

Somewhere in the cleaner part of the world, KPMG put out a report called Redefining excellence in the age of agentic AI. You can smell the title from three blocks away. It has the perfume of conference carpeting and coffee that died in a silver urn. It sounds like something written by a committee that has never had to unclog a toilet or apologize to a child.

Then the report disappeared.

Not because it was too honest. Not because it burned too hot. It disappeared because organizations named in it started saying, hold on, we never did that. UBS said the claims about its AI use were wrong. The National Health Service said the claims about it were wrong. Swiss Federal Railways, Transport for London, same story. The polished report about the bright corporate future had apparently invented parts of that future, like a drunk at last call describing his boxing career.

An AI detection outfit said the mistakes came from hallucinations.

There is a nice little circle of hell there. A professional services firm uses AI to help write a report about AI, and the AI makes things up about the people supposedly using AI. Then the humans, those expensive mammals with badges and guidelines and perhaps very good shoes, let it walk out the door wearing a tie.

If you wrote that into a novel, an editor would circle it and write: too obvious.

But life has never had good taste.

The wonderful phrase, of course, is “human oversight.” Every company in the AI gold rush keeps that phrase on hand like a priest keeps holy water. Do not worry, they say. There will be human oversight. The machine will accelerate the work, but humans will remain in the loop. The loop will be responsible. The loop will be robust. The loop will wear a lanyard and attend the webinar.

Then the loop goes to lunch.

I do not say this because I think humans are noble angels tragically betrayed by silicon. Humans lie, guess, fake, pad, forget, skim, and sign things they have not read. I have known men who would certify a burning building as safe if quitting time was close enough. The machine did not invent laziness. It just gave laziness a faster horse.

That is what scares me about this stuff.

Not that the machine hallucinates. People hallucinate all the time. We call it memory, ambition, dating, politics, quarterly planning. The dangerous part is that the machine hallucinates in the voice of authority. It does not mumble from a barstool. It speaks in complete sentences. It formats the lie. It puts the lie in a table. It cites things with the confidence of a district attorney and the soul of a slot machine.

And the suits love complete sentences.

A lie in a dirty shirt has to work hard. A lie in a report template gets invited upstairs.

That is the whole racket of professional authority. The margin is clean, the logo is expensive, the prose has been sanded until no fingerprints remain, and everyone pretends the words must have passed through some adult chamber of judgment. There is supposed to be a person somewhere. A careful person. A tired but honorable person with a red pen, a phone, and enough shame to verify whether the thing being said actually happened.

Maybe that person used to exist.

Maybe he retired.

Maybe she got laid off during a transformation initiative.

Maybe she now spends her days prompting the machine to produce twenty versions of the sentence “AI is reshaping the enterprise landscape” while her own landscape collapses into tabs and calendar invites.

KPMG says people are expected to follow guidelines on responsible AI use, including oversight and source checking. Good. I am also expected to floss. Expectations are beautiful little paper boats. Reality is the gutter after rain.

The problem is not that guidelines do not exist. The problem is that guidelines are often where responsibility goes to die in a PDF. Everybody points to them afterward. See, the rule was there. The rule had bullet points. The rule had a title and a version number. Therefore the organization cared.

No.

Caring is when somebody with authority slows the machine down.

Caring is when the report is late because a human being picked up the phone and asked UBS whether UBS actually did the thing the shiny paragraph says UBS did.

Caring is when the client gets annoyed, the partner gets twitchy, the deadline makes animal noises in the hallway, and someone still says: we are not publishing this until we know it is true.

That costs money.

That costs time.

That costs the one thing most modern institutions hate spending: attention.

At the post office, we had supervisors who loved procedures. They had binders full of them. The binders had tabs. The tabs had labels. The labels had the dead seriousness of a government tombstone. But when the mail backed up and the machines jammed and the floor turned ugly, the procedure became a decorative object. What mattered was whether somebody knew what the hell they were doing and whether anybody listened to him before the bags piled up to the ceiling.

Usually they listened afterward.

Afterward is the most popular tense in management.

After the report is published. After the fake citations are found. After the client calls. After the journalists start asking. After the PDF is pulled down and the spokesperson releases a sentence that has been washed so clean it no longer contains blood.

KPMG is not alone. That almost makes it worse. EY pulled a report with fake footnotes. South Africa had to withdraw an AI policy after fabricated academic citations crawled out of the walls. The same story keeps putting on different jackets. A big institution asks the machine for authority. The machine provides the shape of authority. The institution forgets that shape is not substance. Then everybody looks surprised when the cardboard bridge gets wet.

We are building a world where official nonsense can be manufactured at industrial speed.

That is not a small problem.

A bad blog post is one thing. A fake case study in a consulting report is another. A false medical claim, a bogus policy citation, a made-up legal precedent, a risk assessment padded with ghost facts—these things do not stay inside the document. They move. They get quoted. They become slides. They enter meetings. They become decisions. They acquire children.

The lie gets a badge.

And because the first version looked professional, each copy looks a little more real.

Like that copy machine in the corner, coughing and flashing green, laying the same black streak down the page while the kid in the blue shirt kept stacking evidence of failure into a neat pile.

I wanted to tell him to stop.

I did not.

Maybe I was tired. Maybe the line behind him was too long. Maybe I have become another citizen of the kingdom of visible defects.

The machine kept working.

The copies kept coming.

And every page looked official enough to ruin somebody’s morning.

Source: KPMG pulled its AI report after UBS, the NHS, and others said its claims about them were made up