Accuracy & truth: 20 ways AI gets it wrong
Hallucinated facts, invented statistics and stale information stated as current. Each failure mode below is phrased as the question people actually ask, with what it looks like in real work — and the layer of the Trust Stack that catches it.
Can AI make up facts that sound completely true?
A briefing states a fact with full confidence that turns out to have no basis in reality.
Does AI invent statistics that were never published?
A report quotes a precise percentage that appears in no real dataset.
Why does AI invent legal rules that don't exist?
A memo describes a rule of law that no statute or court ever established.
Can AI fabricate regulations that aren't real?
A compliance note cites a regulation that the regulator never issued.
Why does AI cite court cases that were never decided?
A contract memo cites a precedent that was overturned — or never existed.
Does AI make up details about real companies?
A profile lists revenue, leadership, or history for a firm that none of it matches.
Can AI fabricate market data and figures?
A market summary cites a market size or growth rate with no real source behind it.
Does AI invent product features that don't exist?
A comparison credits a product with capabilities it has never had.
Why does AI invent people and job titles?
An email is addressed to a contact whose name and role were never confirmed.
Does AI mix outdated facts with current ones?
An answer blends last year's pricing with this year's policy as if both were live.
Can AI give information that's no longer true?
A how-to guide relies on a process that changed months ago.
Why does AI present assumptions as established facts?
A plan states a budget figure as fixed when it was only an unverified guess.
Does AI sound certain when it should be unsure?
A risky claim is delivered in absolute terms with no hedging at all.
Can AI use vague wording to hide thin evidence?
Phrases like "studies show" stand in for any actual cited study.
Why do AI explanations sound right but be wrong?
A cause-and-effect story reads cleanly but rests on a false mechanism.
Can AI reach the right answer for the wrong reason?
A correct conclusion is justified by reasoning that doesn't actually support it.
Does AI overgeneralize from too little information?
A single example becomes a sweeping rule applied to every case.
Why does AI misread what the user actually wants?
A polished deliverable solves a different problem than the one that was asked.
Does AI miss the important exceptions to a rule?
A general rule is applied to a case that is explicitly carved out from it.
Can AI confuse what the law says with what people do?
Common practice and personal opinion are presented with the same authority as binding law.
More from the library
One model can’t reliably catch its own mistakes. A council of independent minds can.
Run your work through the councilAll 250 failure modes · See also: the Trust Stack