The Schwartz Incident

In The You On AI Encyclopedia

The incident is instructive not because it is typical but because it is the visible edge of a phenomenon that is usually invisible. The fabricated citations were caught because they did not exist — the failure mode was binary, detectable through straightforward verification that anyone could perform.

The more consequential gap between competence and comprehension produces failures that are not fabrications but distortions: real citations misrepresented in subtle ways, real cases cited without the qualifications that reshape their meaning, real arguments structured in ways that overlook the counter-arguments a thorough reader would have anticipated. These failures are not caught by verification; they are caught only by the practitioner who has read the cases with the depth the formal standard of legal practice requires.

Diane Vaughan

The Schwartz incident operates in Vaughan's framework as an inverse demonstration. The failure was detected because Schwartz's reliance on the tool produced a binary error the existing verification processes could catch. A lawyer who used the tool more competently — who checked that cited cases existed but did not read them with the depth that would reveal the subtle distinctions, qualifications, and counter-arguments — would produce outputs that pass every verification check while occupying the comprehension gap the formal standard of practice was designed to prevent.

The incident produced a wave of bar association guidance, firm policies, and technology vendor disclaimers addressing AI-fabricated citations. None of this guidance addresses the deeper gap. The policies require that citations be verified; they do not require that cases be read. The comprehension gap persists in the practice even as the binary failure mode has been largely eliminated.

Origin

The case was Mata v. Avianca, Inc., filed in the Southern District of New York. The sanction order was issued by Judge P. Kevin Castel on June 22, 2023. Schwartz and his firm were fined $5,000. The incident was widely covered in legal and technology press and became a standard reference in discussions of AI use in professional practice.

Key Ideas

Binary failure caught. The citations were fabrications; verification caught them because the cases did not exist at all.

Marginal failure survives. Outputs that cite real cases but misrepresent their meaning pass verification while occupying the comprehension gap.

Policy response insufficient. Bar guidance requires citation verification but does not require reading; the gap persists in practice.

Visible edge of invisible phenomenon. Most AI-era comprehension failures are not binary and will not be caught by the mechanisms that caught Schwartz.

Reinforcement of normalization. The attention paid to binary fabrications may paradoxically reinforce the sense that verified outputs are trustworthy, accelerating the drift in comprehension standards.