The Evidentiary Standard Has Changed
The cases that established the precedent.
Courts across four jurisdictions have treated the absence of a contemporaneous verification record as evidence of fault — independent of whether an error was made. These are the primary-source cases that documented the pattern.
Miller v. Regions Bank · N.D. Alabama · Case No. 2024-cv-01324 · 2026
The Two-Track Doctrine
AI hallucinations in a federal brief. Judge Mooty applied a two-track sanctions doctrine: the error itself — submitting a brief containing fabricated citations generated by ChatGPT — received a public reprimand. But the destruction of the ChatGPT account after the court's order — eliminating all contemporaneous record of what occurred and when — received adverse inference, contempt findings, and a six-month suspension from practice.
The distinction was explicit: the AI error was a professional mistake. The destruction of the record was evidence of fault.
Six-month suspension. The sanction was for the missing record, not the error. Source: PACER.
H.C. v. Contreras · California 5th DCA · F089316 · 2026
Three Gates. Zero Records.
A fabricated citation was flagged by opposing counsel during the proceedings. The trial judge adopted the citation anyway, copying it verbatim into the ruling — including an identical typographical error that proved it came from the same AI-generated source. Three separate verification gates — the drafting attorney, the reviewing attorney, and the judge — all failed. No contemporaneous record of review existed at any stage.
The appellate court identified the pattern through the typographical error. The case was not merely reversed — it was reassigned to a different judge entirely.
Reversed and reassigned to a new judge. Source: courts.ca.gov.
CPS / Tobosaru · EWHC 1720 (Admin) · United Kingdom · 2026
The State as Actor.
The Crown Prosecution Service — the state prosecution authority of England and Wales — submitted hallucinated case citations to the High Court in an extradition appeal. The fabricated material migrated from one internal CPS document to another, surviving the permission stage unchallenged. Each reader in the chain assumed that someone upstream had already verified the citations. No one had.
This case extended the pattern beyond private practice: the state itself was the actor.
CPS admitted: "human error in failure to verify." Source: BAILII.
Law Society of Ontario · Mary Hyun-Sook Lee · Canada · 2026
First Canadian Suspension for AI Misconduct.
A lawyer submitted ChatGPT-fabricated citations in an estate matter before the Ontario courts. When questioned directly by the judge about whether AI had been used, the lawyer denied it. The Law Society Tribunal was explicit in its reasoning: the dishonesty — the absence of any truthful contemporaneous account of what had occurred — was what drove the severity of the sanction.
This is believed to be the first time a Canadian legal regulator has suspended a lawyer specifically for AI-related misconduct.
6-month suspension · $10,000 costs · Criminal contempt referral to Attorney General. Source: Law Society of Ontario.
Regulatory Context
ITALY · 2025 (in force)
Legge 132/2025, Art. 13
AI use in intellectual professions must be instrumental, with human judgment prevailing. The professional bears full responsibility for AI-assisted output and must inform the client of AI use. Documenting the verification process is the recommended way to demonstrate the due diligence the law demands. In force since October 2025.
ZTR provides the contemporaneous record that demonstrates the due diligence the law demands.
EU · 2 August 2026
EU AI Act — Reg. 2024/1689
Transparency obligations (Art. 50), market surveillance (Chapter IX), and sanctions become enforceable on 2 August 2026. High-risk system requirements including human oversight (Art. 14) are deferred to December 2027 / August 2028 by the Digital Omnibus (Reg. EU 2026/1744).
ZTR provides the demonstrability layer for human oversight that the regulation requires but cannot itself produce.
ITALY · 29 July 2026
CCNL Sanità 2025-2027
First Italian collective agreement to include a full title dedicated to AI. Covers 592,638 healthcare workers. Establishes that decision-making responsibility cannot be delegated to algorithms. Requires prior disclosure to employees on AI systems, their purposes, data processed, and impact on work organisation.
ZTR extends beyond legal to healthcare — the same verification record, applied to clinical and administrative review where AI assists decision-making.
USA · 2024
ABA Formal Opinion 512
Lawyers using AI must apply competence (Rule 1.1) and supervision (Rule 5.1) standards. Every AI-assisted filing must be independently verified before submission.
ZTR creates the contemporaneous record of that verification.
USA · 2026
Rhode Island Rule 1.1 Amendment
The Supreme Court amended the competence rule to explicitly include understanding AI capabilities and limitations. Independent verification before filing is now a stated requirement.
ZTR documents compliance with the verification requirement.
UK · 2026
Nuffield Foundation · Sargeant Policy Brief
Of 45 AI tools deployed in the England and Wales justice system, only 7 (15.5%) have been independently evaluated. Human oversight is treated as the principal safeguard — yet no deployment evaluation examines whether it operates effectively in practice.
ZTR is the first instrument purpose-built to make human oversight demonstrable at the individual document level.