AI Failure Patterns
Public AI failures leaders can govern before they escalate.
Public AI failures are not random. They cluster around recurring breakdowns in evidence, context, reasoning, competence, and judgment. WEKID™ turns those patterns into a practical governance lens.
This catalogue maps 45 public cases across 24 sectors to the WEKID layers — Data, Information, Knowledge, Experience, and Wisdom. The purpose is not to claim any framework eliminates risk. It is to show where governance should detect, constrain, escalate, or mitigate risk before harm occurs.
Executive summary
AI failures repeat because organizations confuse output quality with decision authority.
A fluent answer can still be unsupported. A correct fact can still be unsafe. A useful agent can still lack operational competence. WEKID separates those questions so organizations can govern AI before outputs become decisions and decisions become incidents.
Pattern across the catalogue
A WEKID Observation Across Sectors
What becomes striking when these cases are viewed together is that the failures do not distribute randomly. They cluster around the same recurring mistakes:
Data failures
- Hallucinations
- Deepfakes
- False citations
- Incorrect records
Information failures
- Missing context
- Alert fatigue
- Moderation ambiguity
- Poor signal quality
Knowledge failures
- Biased inferences
- Incorrect generalizations
- Misapplied rules
Experience failures
- Premature automation
- Procedural incompetence
- Failure under edge conditions
Wisdom failures
- Delegating authority beyond competence
- Treating probabilities as judgments
- Removing meaningful human oversight
- Prioritizing efficiency over legitimacy
A sample case panel
MASKED · 2023 12
Fabricated Legal Citations in Court Filings 3
Hallucinated facts treated as authoritative truth. 4
DataWisdom 5
Legal · Rejected 6
How to read each case
Every panel answers six questions.
- Identity badge. The gold MASKED mark stands in for a real, publicly documented organization whose name is withheld in the public catalogue.
- Period. The date marks when the incident occurred — the year (or range) the case is associated with.
- Case. What happened, stated neutrally — described by the failure, not framed as an accusation.
- Failure pattern. The underlying epistemic mistake in WEKID terms — the question the system answered before it had earned the authority to.
- WEKID layer(s). Where the failure originated — Data, Information, Knowledge, Experience, or Wisdom (color-coded; some cases span more than one).
- Sector & outcome. The domain, and the WEKID Decision-Matrix verdict the framework would reach: Approved, Monitoring, Constrained, Remediation, or Rejected.
MASKED
Why are names hidden? The public catalogue is masked so the focus stays on the governance lesson, not the organization. The full named organizations, individuals, sources, and timelines behind every case are released in the WEKID Executive Brief.
Unlock the full record →| Case | WEKID layer | Sector | Period | Outcome |
|---|
The emerging thesis
The Emerging Thesis
If you were to summarize all of these examples in a single sentence for the WEKID trilogy, it would be this:
Across healthcare, finance, education, defense, government, transportation, and commercial enterprise, the defining AI failures of the modern era have not primarily been failures of intelligence. They have been failures of governance — instances in which systems were granted authority beyond the quality of evidence, context, experience, and judgment available to support their decisions.
That observation may ultimately be the strongest empirical validation of the WEKID framework: regardless of industry, the same epistemic failure patterns emerge again and again.
turn failure patterns into an adoption roadmap
Use these patterns to assess your AI governance exposure.
WEKID adoption starts by identifying where your AI systems operate today, where stakeholders believe they operate, and where authority should actually reside.