No, AI Is Not About to Replace Lawyers

Of all the AI replacement scenarios that float around in business press and Twitter threads, the claim that artificial intelligence is about to make lawyers obsolete is among the most persistent. It surfaces almost monthly, usually attached to a new survey, a courtroom-related lawsuit, or a viral demo of a model drafting a brief. The argument sounds plausible on its face. Law is text-heavy, expensive, and structured. Surely text-prediction software has it cornered.

The actual evidence says otherwise, and it has done so consistently across studies and surveys throughout the past year.

What current models can and cannot do

The most-cited recent piece of research comes from Stanford's RegLab, which tested headline large language models on legal questions and found they hallucinated on 58 to 82 percent of prompts depending on the model. Even bespoke legal-research tools, the ones lawyers actually pay for and use day to day, returned a wrong answer roughly once every six queries. A model that invents a precedent one time in six is not one a working attorney can lean on without verifying every cited authority by hand. The Stanford team's conclusion, reported across Science and others, was that headline accuracy figures from the vendors badly overstate real-world reliability.

The Wolters Kluwer 2026 Future Ready Lawyer report puts the addressable share of a lawyer's working hours at around 23 percent. That is a meaningful figure. Document review, contract drafting, and first-pass legal research are the obvious targets, and large firms are already restructuring around them. But 23 percent automation of tasks is not the same as replacement of the worker doing them. A lawyer who spends 23 percent less time on review still has to handle the other 77 percent of the work, and the lawyer is the only one allowed to.

What the replacement narrative ignores

The argument usually leans on a flat picture of lawyering, as if the work were essentially document production with a few courtroom appearances bolted on top. The reality, articulated by every senior practitioner who has been asked, is that the durable parts of the job are the ones AI cannot do. Reading a client's actual problem, often quite different from the one they describe. Judging which arguments will land in front of which judge. Negotiating against another human being with their own incentives and stress responses. Carrying ethical weight that the model is not legally permitted to carry on anyone's behalf.

That last point matters. The New York State Bar Association's 2026 guidance frames it bluntly: AI can inform the work of a lawyer, but it cannot substitute for one. The American Bar Association has reached a similar position in its model rules, which now explicitly require attorney supervision of any generative-AI output used in client work. Courts have already begun sanctioning lawyers who treated AI output as a substitute for verification, with fines running into six figures in some 2026 cases.

Where the change is actually happening

The interesting question is not whether lawyers will be replaced but how the profession is redistributing itself around the new tools. Junior associate hours billed for document review are dropping, sometimes sharply. Specialised AI-fluent firms are charging less for routine work and more for advisory. Compliance departments inside companies are growing in headcount, not shrinking, because someone has to design and audit the AI tools the company now relies on. None of this looks like the profession disappearing. It looks like a profession that has absorbed several new technologies before, and is doing it again.

Spell-check did not eliminate lawyers. Westlaw did not. The likeliest outcome with generative AI is that it follows the same path. The lawyers who learn the tools well will price out the lawyers who refuse to touch them. The legal labour market shifts in shape rather than shrinking in size. That is a real story, and probably the one worth telling. "Lawyers are obsolete" is not.