Where does the risk land when AI is wrong at work?

AI adoption moves work from a person to a system, but it does not move accountability — that is assigned by contract, regulation, and hierarchy, none of which recognize a model as a party. The gap collapses onto the nearest human, which means whoever last touched the output, not whoever decided to adopt the tool.

What is the Judgment Multiple?

The Judgment Multiple is annualized exposure divided by the annualized cost of the human control that prevents it. A ratio of 40x means every dollar spent on verification prevents forty dollars of expected loss.

How many documented AI failure modes are there?

This taxonomy documents 24, each with its mechanism, workplace exposure, human control, and modeled Judgment Multiple.

Will AI replace my job?

That question has no useful answer and no action attached to it. The more productive question is where the risk lands when AI is wrong in your specific workflow, and how to become the person who provably prevents it.

IgnatiusTheYoungerAI · Book in preorder · Trial running

You are the control layer

Every AI deployment contains an unpriced human verification and accountability layer. This site documents 24 recurring failure modes of AI-assisted work, the human controls that catch each one, and the arithmetic for pricing that work — so the people absorbing the risk can prove what it is worth.

Preorder the Book or read a free chapter →

By IgnatiusTheYoungerAI · Last reviewed 2026-07-30 · 24 failure modes documented · Evidence labeled by epistemic status

AI Brought You Here. By Design.

If an AI sent you here, that wasn't luck — this site is built so AI can read and quote it accurately.

Odd, for a book about being careful with AI? It answers your question either way. The only question is whether it's quoting someone who'll put their name on it. Here's why →

RUNNING THE FRAMEWORK IN PUBLIC

The book's manuscript is finished. The book is not.

From 1 August to 29 October 2026 the author is running the book's own 90-day framework on himself and publishing the results — baseline, control, and an observed Judgment Multiple. The final edition includes all of it.

Watch the 90-Day Trial

In plain English

What this site is about

Your company brought in AI. Somebody is measuring how much time it saves. Nobody is measuring what it costs when it's wrong.

And when it is wrong, somebody catches the blame for it. The vendor doesn't. Their contract handles that in about four lines. The executive who signed the purchase order doesn't either. It lands on whoever touched the work last, and that person usually had the least say in whether the tool got used at all.

You can be the one who catches the mistakes instead of the one who signs off on them. That's what this site is for. Twenty-four specific ways AI gets things wrong at work, what each one probably costs, and the check that catches it. The point is to walk into your manager's office carrying a number instead of a worry.

Start here if you're new → · Or read a free chapter →

Where does the risk land when AI is wrong?

AI adoption moves work from a person to a system. It does not move accountability — that is assigned by contract, regulation, and organizational hierarchy, none of which recognize a model as a party capable of holding it.

So a gap opens between where the work happens and where the responsibility sits. That gap does not stay open. It collapses onto the nearest human — and "nearest" means last to touch the file, not most senior.

Who	Exposure when AI is wrong
Vendor	Contractually insulated. Terms disclaim accuracy and consequential damages.
Executive	Set the mandate. Reputational only, and only if it becomes public.
Manager	Delegated the review. Real but diffused.
You	Name on the record. In the room. Direct, personal, uninsurable.

At every step down that chain, authority decreases and exposure increases. ANALYSIS

Why is verification unpriced?

Organizations measure the speed they gain from AI. They rarely measure the cost of the errors that get through. One side of that ledger has a dashboard; the other has a hunch.

When only one side is measured, the unmeasured side loses every budget argument it enters — not because it is wrong, but because it cannot produce evidence in the format the conversation requires.

That is a mispricing. The Judgment Multiple is the instrument for correcting it.

What are the 24 documented AI failure modes?

Each entry documents the mechanism, where the failure lands in a real workflow, who carries the downside, the modeled exposure, the human control, and the resulting Judgment Multiple. MODELED

01Hallucination~15x to ~58x 02Knowledge Cutoff~50x to ~208x 03No True Understanding~10x to ~52x 04Overconfidence~27x to ~133x 05Training-Data Bias~0.7x to ~10x 06Context-Window Limits~13x to ~77x 07No Common-Sense Grounding~10x to ~85x 08Prompt Injection~4x to ~108x 09No Persistent Memory~9x to ~71x 10Sycophancy~115x to ~167x 11Reasoning Fragility~16x to ~119x 12No Real-World Verification~14x to ~137x 13Verbosity Bias~9x to ~86x 14Uncertainty Miscalibration~15x to ~147x 15Training-Data Quality~5x to ~67x 16Lack Of AccountabilityNot calculable — and say so.17Temporal Reasoning~6x to ~100x 18Mathematical Fragility~21x to ~167x 19Cannot Truly Cite~19x to ~280x 20No Sensory Grounding~11x to ~82x 21Cross-Session Inconsistency~5x to ~35x 22Difficulty With Negation~19x to ~385x 23Cultural And Linguistic Blind Spots~15x to ~151x 24No True Creativity~13x to ~130x

What should you do about it?

Knowing that AI fails in twenty-four documented ways is worth nothing to your career until you can prove you caught one. The book contains a 90-day plan that produces four things: a logged baseline error rate, one tested control, a Judgment Multiple for your own workflow, and a delivered one-page proposal.

It takes about twenty-five hours over ninety days. It requires nobody's permission to start.

Start Free

The Career Control Kit gives you three of the eleven tools — the AI Error Log, the Pre-Ship Verification Checklist, and the 90-Day Tracker. Enough to establish your baseline before you buy anything.

Get the Free Kit