Mandatory independent bias audit protocols and statutory employment agency liability for third-party AI software screening job applicants on video, voice, and resume models.
↳ Statutory Hook: NYC Local Law 144 & Title VII (42 U.S.C. § 2000e-2)Foundational Supreme Court standard governing the admissibility of scientific, algorithmic, and forensic expert evidence in federal courts.
Federal Rule of Evidence 702 requires trial judges to act as gatekeepers ensuring that scientific and technical expert evidence is both methodologically reliable and relevant.
Demands transparent error rate benchmarks, reproducible test datasets, and formal algorithmic validation before AI outputs or forensic models can be admitted into evidence.
Binding SCOTUS Evidentiary Bedrock
“The Frye 'general acceptance' test was superseded by the Federal Rules of Evidence. Under Federal Rule of Evidence 702, the trial judge must ensure that all scientific and technological testimony is not only relevant, but reliable, acting as a judicial 'gatekeeper'.”
Plaintiffs Jason Daubert and Eric Schuller were born with severe limb-reduction birth defects. They alleged the birth defects were caused by their mothers' ingestion of Bendectin, an anti-nausea drug marketed by Merrell Dow. The trial court excluded plaintiffs' expert witnesses because their statistical animal and re-analysis studies were not published or generally accepted in the epidemiological community.
On writ of certiorari to the Ninth Circuit, which had affirmed summary judgment excluding plaintiffs' expert epidemiological testimony under the 70-year-old Frye 'general acceptance' standard.
Issue: Whether the 'general acceptance' test established in Frye v. United States was superseded by the adoption of the Federal Rules of Evidence (specifically Rule 702) for determining the admissibility of scientific and technological expert evidence.
Daubert is the universal federal framework governing the admissibility of all complex technological evidence, computer modeling, and artificial intelligence forensics. When litigants seek to introduce synthetic media detection, probabilistic machine learning outputs, automated facial recognition, or algorithmic risk assessments, Daubert requires courts to evaluate testability, error rates, and algorithmic validation.
Daubert v. Merrell Dow Pharms., Inc., 509 U.S. 579 (1993).