For the first time anywhere in the world, a UK based AI law firm beat a traditionally represented opponent at trial. The implications reach every common law jurisdiction
The defendant had a solicitor and a barrister. Garfield AI had a chatbot and a recently called junior advocate. On 22 June, at Wandsworth County Court in southwest London, Garfield won.
The case wasn’t huge – it was a £7,000 ($16,000) dispute over unpaid freelance fees. Garfield AI, a London firm that prepares legal claims using artificial intelligence rather than human lawyers, drafted all the pre-trial documents - correspondence, claim forms, witness statements - and brought in a human barrister for the advocacy alone. The client, freelance HR consultant Tamires Camal Taquidir, paid approximately £400 and recovered the full £7,000. The other side paid for both a solicitor and a barrister and lost.
Founder Philip Young, a former City of London litigator, called it "a landmark moment, not just for Garfield AI, but for access to justice."
Garfield received approval from the Solicitors Regulation Authority in May 2025 - the first and still only AI-only firm licensed to provide regulated legal services anywhere. Nothing equivalent exists in New Zealand. The New Zealand Law Society regulates the provision of legal services under the Lawyers and Conveyancers Act 2006, and the Act requires that regulated services be provided by a person holding a current practising certificate. An AI system cannot hold one.
That settles the immediate question. The harder one is what the profession does with AI in the meantime. New Zealand firms - from suburban sole practitioners to the larger city practices - have been absorbing AI tools into daily work for the past two years, as NZ Lawyer's legal technology coverage has tracked. The Garfield result shifts the terms of that conversation. Until last week, AI in legal practice meant AI assisting a lawyer. Now there is a regulated precedent, confirmed by a court, for AI replacing the preparatory work a lawyer would have done - and winning.
The trial win came alongside a fresh pair of cautionary examples. Sullivan & Cromwell - one of Wall Street's most prominent firms - disclosed to a US federal bankruptcy court last week that a recent filing contained multiple AI-generated errors. Pinsent Masons was publicly reprimanded by a London court in May for submitting AI-drafted letters that twice misquoted insolvency legislation. The New Zealand Law Society has not issued dedicated AI practice rules, but the duties of competence and supervision under the Lawyers and Conveyancers Act apply to AI-assisted work as they do to any other. A court will not accept "the AI got it wrong" as an answer.
Garfield's design is relevant here. The SRA required the firm to disable case law citation - identified as the highest-risk area for large language model errors - and named human solicitors remain professionally accountable for the firm's output. Those constraints are part of why the firm passed regulatory scrutiny and, likely, part of why its documents held up in court.
Globally, the pace is not slowing. Kirkland & Ellis, the world's highest-grossing law firm, announced last month it is committing US$500m to a proprietary AI platform. Freshfields signed a deal with Anthropic in April. New Zealand firms are at an earlier point on that curve, but the economics that drive the investment - the same economics that let Taquidir recover £7,000 for £400 - are not jurisdiction-specific.
Access to justice is a live issue in New Zealand civil practice. Disputes that are real but modest routinely go unpursued because legal costs exceed the amount at stake. A model that can prepare a complete court file for under £400, with a human appearing for the hearing, addresses that problem directly. The New Zealand regulatory framework does not currently permit it. Whether that changes, and how, is a question the profession will need to engage with - the Garfield verdict just moved it further up the agenda.
