Around 370 BC, Plato has Socrates tell the story of an Egyptian god who invents writing and presents it to the king as a gift for memory. The king refuses it. People who rely on letters, he says, "will appear to be omniscient and will generally know nothing; they will be tiresome company, having the show of wisdom without the reality." That is the fake-expert charge, word for word, aimed at the written word. Twenty-four centuries later we are aiming it at a chatbot, and it is doing what it has always done: slowing the people who use the new tool down, but ultimately failing to stop them.
Three reasons to keep quiet
In August we wrote that we built the perfect conditions for hiding AI use. The numbers since have only filled in the picture. Most people using AI at work are hiding it, and they have three good reasons.
There is no rule to follow. In the KPMG and Melbourne study, fewer than half of employees had received any AI training, and 48% admitted using AI in ways that broke company policy. When the policy is silent, or a blanket ban nobody believes in, the safest move is to use the tool and say nothing. Silence can't be held against you. A question can.
Your peers will think less of you. This one is measured now, not assumed. Jessica Reif, Richard Larrick and Jack Soll at Duke's Fuqua School ran four preregistered experiments and found that people who use AI are rated lazier, less competent and less diligent than people who get the same help from a colleague or do the work unaided. Participants also predicted the penalty before they received it. Everyone knows the rule, and it gets enforced anyway.
It makes your work look cheaper. Fifty-three percent of AI users in Microsoft's survey worried that using it on important work makes them look replaceable. If the draft took ten minutes, why did the invoice say four hours? The fear is rational. Much of professional pricing is a claim about effort, and AI breaks the link between effort and output.
The crutch accusation is older than you
The story usually told about this is generational. Older professionals, who learned the craft the slow way, look at younger ones leaning on a machine and see softness: a generation that needs a crutch, that can produce the answer without understanding it, that is performing expertise instead of having it. Millennials heard it about Google and Wikipedia. Gen Z hears it about ChatGPT. The ones now in their forties are frequently the ones saying it.
The accusation has some truth in it. You can use AI to fake competence, and some people do. But look at the pattern rather than the grievance. The same charge met calculators, spell-checkers, search engines and, if you believe Plato, the alphabet. Each time, the old guard was defending a real skill and a real status, and each time the status went first.
The Duke data adds a twist the generational story doesn't survive: the social penalty for AI use was uniform across age. Young colleagues mark each other down too. The crutch accusation belongs to a status system, and anyone with a stake in it will use it, whatever their birth year. That is why it won't fade just because the people who invented it retire.
Call the cost of all this the Crutch Tax: the time, money and candour a professional spends to look unassisted. It is paid in hours logged for work a model did in minutes. It is paid in the junior analyst who can't tell her manager which half of the report she checked by hand, so neither half gets credit. It is paid by the client, who is billed for the performance of effort.
Prestige doesn't get fixed. It gets overrun
I would like to tell you the prestige question gets resolved, that professions will hold a conversation, update their norms and decide AI-assisted work is respectable. That is the slow road, and it is not the one that matters. The machinery being threatened here, billable hours, apprenticeship ladders, the idea that expertise shows in how long something takes, has been in place for a century. It will not be unwound by persuasion in a few years.
It doesn't need to be. It will be undercut.
Law gives us an early look. Clio's 2026 report on solo and small firms found that three quarters now use AI, yet fewer than a third report any revenue increase from it, against nearly 60% of enterprise firms. The work got faster and the billing did not follow. Somebody is capturing those saved hours, and increasingly it is the client, who notices that the firm across town turns a contract around in a day for less money.
That is how the argument ends. Nobody wins the debate about whether AI-assisted work is real expertise. The debate simply stops mattering once a competitor offers the same result for a lower price, or an investor sees one firm returning more per head than another. Markets have little patience for guilds. The bar association can decide what counts as craftsmanship. Your client decides what they are willing to pay for.
So whether you like it or not, if you don't pick up the tool, a competitor will. The only choice left is how you pick it up: in hiding, paying the Crutch Tax, or in the open.
Why hiding loses anyway
Suppose you accept all of that and still prefer discretion. Use it, bill as before, say nothing. For a while it works. It is also the worst position to be in when the market turns, for two reasons.
First, you can't price what you can't admit. The firm that says "we use AI for the first draft, a partner checks every clause, and our fee reflects both" can compete on price and on trust. The firm pretending the work is still hand-made can compete on neither, because the moment it cuts its price it has confessed.
Second, you can't protect what you won't disclose. The same KPMG study found 57% of employees rely on AI output without checking its accuracy. Hidden use is unreviewed use by default, because nobody reviews a process they have been told doesn't exist. And the documents flowing into those tools, client contracts, patient notes, HR files, are flowing with no control on them at all. We covered what that looks like at a ten-person firm: nobody finds out until the client does.
What disclosure looks like
Disclosure means stating how the work was made, plainly, in the same breath as the work. Nobody is asking for a confession. It has three parts.
Say where the machine was. "AI drafted the summary; I rewrote the recommendation." One line, on the work itself. The goal is that nobody has to guess.
Say who checked it. Accountability is the thing the profession was defending all along. Every publicised AI failure in law, the invented citations and imaginary precedents, was sanctioned as unreviewed work. A named person who read it and stands behind it answers that objection completely.
Say what left the building. Once use is in the open, the useful question becomes what data the tool saw. That one has a practical answer: strip names and identifiers from the document before it goes to the model, and restore them locally afterwards. It's the reason we built promptShield, and it only works for people who admit they are using AI in the first place.
We had to take our own advice. Every post on this blog carries a byline that reads "drafted with Claude, reviewed and edited by hand." We argued about the word order. An earlier version put the model first and the human second, and we rejected it, partly because it misstated who is accountable and partly, if we're being candid, because it looked worse. That second reason is the Crutch Tax, paid by the people writing an article about the Crutch Tax. We kept the order because the first reason holds. We're telling you about the second because it's the point of this piece.
The age of disclosure
Nobody will announce the change. There will be no memo declaring AI-assisted work respectable. The shift will show up the way these shifts always do: in a quote that comes in lower, in a proposal that lists its method, in a job ad that asks which tools you use and how you check them.
What the change does to people is harder. The junior staff who learned the craft by doing the grinding first draft will need a new ladder, and the firms that hide their AI use are the least likely to build one, because they can't admit the old ladder is gone. The professionals who hid will have spent years paying a tax to protect a status that was going to be repriced regardless.
You don't have to wait for your profession to make peace with this. Tell one colleague how you actually work this week. Put the method next to the output. The competition is going to make disclosure normal either way. You can be early to it, or you can be caught by it.
Common questions
Why do people hide their AI use at work?
Three reasons recur in the research: most workplaces have no clear AI policy, colleagues judge AI users as lazier and less competent, and people fear AI makes their work look cheaper or their role replaceable. In one 47-country study, 57% of employees said they hide their use.
Is the stigma against AI a generational thing?
Partly in how it is told, much less in the data. Older professionals often frame AI as a crutch for a softer generation, but Duke's experiments found the penalty for AI use was the same across ages. It behaves like a status defence rather than a generational one, which is why retirement alone won't end it.
Should I tell my employer or clients that I use AI?
Yes, with the method attached: where the tool was used, who reviewed the result, and what data it saw. Disclosure is what makes review and data protection possible. Our guide on what happens when you paste documents into ChatGPT covers the data half.