The Citation Failure Record

The Institute's charter commits it to keep a public record of AI citation failures in court "so that bar associations, judges, and malpractice carriers can set policy on evidence rather than anecdote." This is that record. Each entry names the case, the court and the date, says what went wrong and what followed, and cites where those facts come from.

What qualifies

An entry qualifies when a court or tribunal has addressed, in a decision, a citation failure in a filing, and the decision or a published source ties the failure to the use of AI. We record only what the cited source states. Where a detail is missing, such as an exact date or the nature of the defect, the field is left empty rather than inferred. A matter is left out altogether until its case name, court and date can be read from a source.

Failure types

Each classified entry carries one failure type.

• Fabricated authority: a cited case, quotation or citation that does not exist.
• Misattributed or misquoted holding: a real authority presented as saying something it does not say, or attributed to the wrong court or case.
• Real case, wrong proposition: a real, accurately quoted case cited for a proposition it does not support.
• Bad law cited as good: authority that was reversed, overruled, superseded or otherwise no longer usable for the proposition, cited as if it were.
• Wrong procedural stage or standard: a holding decided under a different stage, standard, burden or forum, cited for something that depended on it (see Right Law, Wrong Stage).
• Other: a failure the source documents that none of the above describes.

Verification

Every entry cites a primary source, such as the court's order, or a published secondary source, and names the place in the Institute's corpus where the entry is documented. When a secondary source is the only basis, the entry says so, and we replace it with the order once the order has been read. Aggregate figures are quoted exactly as their sources state them, with the source's date and its own definition. The counts measure different things: a database of decisions that address AI use is not a count of fabricated citations.

A seed set, not a census

This first release is small by design. It holds only matters whose details appear in the Institute's own papers and essays, and every classified entry so far is fabricated authority. That skew is a finding about visibility, not frequency. A fabricated citation announces itself: opposing counsel cannot find the case, and the court says so in an order. A real case cited for the wrong proposition, at the wrong stage, or after it stopped being good law announces nothing, which is the argument of Good Law for What? The other failure types are in the taxonomy because the record exists to find them. Proposed entries and corrections are welcome, with the source attached.