Taxonomy · 24 Entries

The 24 Documented AI Failure Modes

A taxonomy of 24 recurring failure modes in AI-assisted work. Each entry documents the mechanism, where the failure lands in a real workflow, who carries the downside, the modeled financial exposure, the human control that catches it, and the resulting Judgment Multiple.

Taxonomy originated by IgnatiusTheYoungerAI · Paired with the Judgment Multiple

Last reviewed 2026-07-30 · Part II of the AI "Keep Your Career" Bible

Every mode follows the same ten beats, so they can be compared directly. Exposure figures are MODELED from stated assumptions and are not measured outcomes from any specific organization.

01HallucinationJudgment Multiple ~15x to ~58x02Knowledge CutoffJudgment Multiple ~50x to ~208x03No True UnderstandingJudgment Multiple ~10x to ~52x04OverconfidenceJudgment Multiple ~27x to ~133x05Training-Data BiasJudgment Multiple ~0.7x to ~10x06Context-Window LimitsJudgment Multiple ~13x to ~77x07No Common-Sense GroundingJudgment Multiple ~10x to ~85x08Prompt InjectionJudgment Multiple ~4x to ~108x09No Persistent MemoryJudgment Multiple ~9x to ~71x10SycophancyJudgment Multiple ~115x to ~167x11Reasoning FragilityJudgment Multiple ~16x to ~119x12No Real-World VerificationJudgment Multiple ~14x to ~137x13Verbosity BiasJudgment Multiple ~9x to ~86x14Uncertainty MiscalibrationJudgment Multiple ~15x to ~147x15Training-Data QualityJudgment Multiple ~5x to ~67x16Lack Of AccountabilityJudgment Multiple Not calculable — and say so.17Temporal ReasoningJudgment Multiple ~6x to ~100x18Mathematical FragilityJudgment Multiple ~21x to ~167x19Cannot Truly CiteJudgment Multiple ~19x to ~280x20No Sensory GroundingJudgment Multiple ~11x to ~82x21Cross-Session InconsistencyJudgment Multiple ~5x to ~35x22Difficulty With NegationJudgment Multiple ~19x to ~385x23Cultural And Linguistic Blind SpotsJudgment Multiple ~15x to ~151x24No True CreativityJudgment Multiple ~13x to ~130x
In plain English

What you're looking at

Twenty-four specific ways AI gets things wrong at work. Each one has a name and its own page. This sits one level below "AI is unreliable," which is true and also completely useless to you on a Tuesday afternoon.

Some are famous: it makes up facts that sound real. Some are sneakier: it agrees with whatever you've already decided, or it drops the word "not" when summarizing a policy.

Each page answers the same questions: what goes wrong, why, where you'd see it in a normal workday, who gets blamed, roughly what it costs, and how to catch it.

You don't need to read all 24. Find the two or three that match your job and start there.

How is this taxonomy organized?

BeatWhat it establishes
The mechanismWhy the system does this — technical, not hand-wavy
Where it shows upOne ordinary workflow. No drama.
Who carries the downsideTraced: vendor → executive → manager → you
The exposureModeled cost with assumptions visible
The human controlWhat a person does, how long it takes, fully loaded
The Judgment MultipleExposure ÷ control cost
Your career moveOne thing to do this week
Evidence noteEpistemic status and sources
The Full Field Guide

All 24 modes in depth, the controls that catch each one, and the 90-day plan for building a documented record that you ran them.

Preorder the Book