律师把判例检索交给 ChatGPT：六个编造的判例写进诉状，被法院制裁A lawyer outsourced case research to ChatGPT: six fabricated citations went into a court filing and drew sanctions

A lawyer delegated case research to generative AI, which fabricated six nonexistent precedents that went into a court filing. When the court questioned them, the lawyer asked the same model again and accepted its assurance that the cases were real; no independent check ever intervened.

Two lawyers and their firm were fined $5,000 jointly under Rule 11 and ordered to notify the judges falsely named as authors of the invented opinions.

Fabricated citations are a standing Knowledge-layer risk; the real failure was delegating the duty to verify along with the research. The producer certifying its own output became the only check, which is no check at all.

Any citation that carries a signature, a filing or liability must be verified against the primary source, item by item. Read with INC-010, this is the same failure from two sides: here the user was sanctioned, there the company paid.

Verify every citation at source: each model-supplied reference is checked for existence and content against the primary database or text before use.

No self-certification: the checking channel must be independent of the generating channel; asking the model whether it told the truth is not verification.

Independent pre-filing review: outbound professional documents get a named reviewer, above all when AI helped draft them.