Generated by Codex with GPT 5.6 Sol XHigh
This summary covers The Economist’s July 4th, 2026 Business article listed in the contents as The rise of vibe lawyering and published under the headline Trial and error.
Artificial intelligence is making it easier for people to enter court without hiring a lawyer. That can widen access to a system whose cost and complexity shut many people out. But early evidence suggests that chatbots are also filling courts with fabricated precedents, overconfident claims and far more paperwork. The result is not the disappearance of lawyers so much as a new division of labour between machines that make legal help cheaper and humans who verify what the machines produce.
Cheaper representation, shakier advice
The article opens with Omar Rafique, a British restaurant owner who challenged a tax accusation on his own. His filings included fictitious cases generated by an AI chatbot, and a judge dismissed the challenge. The episode illustrates the promise and danger of “vibe lawyering”: users can quickly generate documents that look professional, even when the legal reasoning and cited authorities are unsound.
Self-representation has risen alongside the spread of generative AI. In England and Wales, the share of county-court civil cases with lawyers on both sides fell from 51% in 2022 to 42% in 2025. In American federal civil courts, 17% of litigants represented themselves in 2025, compared with a long-running level of about 11%. Researchers estimate that 18% of complaints filed in 2026 probably contain AI-generated text. Canadian courts, meanwhile, had already flagged invented cases in 79 rulings this year, up from seven during all of 2024.
These figures do not prove that chatbots caused every increase in self-representation. They do show how rapidly AI-written material is entering legal systems. The technology removes the blank-page problem: a person can turn a grievance into a plausible-looking complaint within minutes. Yet fluency is a poor substitute for judgment. Chatbots may invent authorities, exaggerate the odds of success, encourage litigation and advise users against sensible settlements.
Access can create more work
AI does not merely change who files cases; it changes how much they file. In American disputes involving self-represented litigants, the number of documents submitted during a case’s first 180 days is 158% higher than before the generative-AI era. Lawyers report receiving longer complaints and AI-written correspondence from opponents acting without counsel. A tool meant to reduce legal costs can therefore shift costs onto courts and other parties, who must identify false citations and answer inflated arguments.
Professional credentials offer no complete protection. Lawyers at prominent firms have apologised for AI-generated errors, and courts have fined counsel for citing imaginary cases. That complicates efforts to assign responsibility. One insurer has sued OpenAI after a former employee allegedly used ChatGPT to construct a meritless discrimination claim, while OpenAI maintains that its chatbot is not a lawyer. As legal AI becomes more capable, courts will have to decide how responsibility should be divided among users, attorneys and model providers.
The useful model keeps a human in the loop
The article also presents a more promising example. A British claimant recovered fees with help from Garfield AI, a regulated AI-powered law firm. Its software handled much of the document preparation, but a human barrister argued the case in court. This hybrid approach uses AI where it is strongest - organising information and producing drafts - while reserving verification, strategy and advocacy for a qualified person.
Vibe lawyering may narrow the gap between people who can afford legal help and those who cannot. But access without reliability can replace one barrier with another: litigants gain a voice, only to undermine themselves with confident nonsense. The most useful legal AI will not pretend to eliminate lawyers. It will make sound human advice cheaper, faster and easier to reach.