Legal AI glossary: the terms that decide cases

These are the terms that recur in the judgments this publication tracks. Each definition is anchored to a decision on the incident tracker where the term did real legal work. This is a seed set; it grows as the record does.

AI hallucination

An output from a generative AI system that is presented as fact but is fabricated — in legal work, a case, quotation, or statutory text that does not exist. Courts stress that the label does not soften the duty: fabricated authority is fabricated authority whoever produced it. See Rafique v HMRC, where the tribunal found the cited cases were “AI hallucinations.”

Fabricated (fictitious) authority

A citation to a case, statute, or rule that was never decided or enacted, or a quotation that appears in no real source. It is the core harm across the tracker, whether produced by AI or otherwise. India’s Supreme Court held that a decision resting even partly on such material is “no decision in the eyes of the law” (Pooja Ramesh Singh).

Verification duty

The professional obligation to confirm that every authority cited is real and says what it is claimed to say, before it is put before a court. In the United States this is grounded in the reasonable-inquiry requirement of Federal Rule of Civil Procedure 11; the Second Circuit referred an attorney to its grievance panel for breaching it in Park v Kim.

Hamid jurisdiction (England & Wales)

The High Court’s supervisory jurisdiction over the conduct of legal representatives who appear before it, derived from R (Hamid) v Secretary of State for the Home Department [2012] EWHC 3070 (Admin). The Divisional Court invoked it to address AI-fabricated citations and refer lawyers to their regulators in Ayinde v Haringey.

Wasted costs order

An order that a lawyer pay costs personally, rather than the client, because of improper, unreasonable, or negligent conduct. Courts increasingly use it for unverified AI citations; in Ayinde, Ritchie J ordered counsel and a law centre to pay £2,000 each at first instance.

Rule 11 (United States)

Rule 11 of the Federal Rules of Civil Procedure requires that filings be grounded in a reasonable inquiry into the law and facts, and authorises sanctions for violations. It was the basis for the first AI-fabrication sanction in Mata v Avianca and for the Wyoming penalties in Wadsworth v Walmart.

Non-delegable duty

The principle that the lawyer who signs a filing owns its contents and cannot shift responsibility to a junior, a contractor, or a tool. Alberta’s Court of Appeal applied it in Reddy v Saroya; Singapore extended it to a supervising solicitor who failed to check a junior’s AI output in Tan Hai Peng.

Abuse of process

Use of the court’s procedures for an improper purpose or in a way that undermines their integrity. Citing fake authorities can meet the threshold: the High Court struck out an appeal on this basis in Bandla v SRA.

Regulatory referral

A court’s referral of a lawyer’s conduct to the body that licenses and disciplines them — the SRA or BSB in England, the Law Society of Ontario, a US state bar, South Africa’s Legal Practice Council, and others. It is the tracker’s distinguishing data point, recorded separately from the court’s own sanction; 21 of the tracked incidents carry one.

AI-use certification order

A standing order requiring parties to certify whether, and how, they used generative AI in a filing. Courts issue them prospectively after an incident; the Delaware Court of Chancery entered one in An v Archblock, warning that non-compliance may lead to sanctions.

Sources

Every term links to a primary-verified decision on the SafeLegalAI incident tracker. Definitions of court rules cite the rule itself: FRCP 11; R (Hamid) v SSHD [2012] EWHC 3070 (Admin).