A senior English barrister has gone public with a dire warning for the legal profession: get out now. Speaking anonymously to The Spectator, the mid-fifties lawyer described an internal experiment where his firm fed a complex civil court appeal to Grok Heavy AI. The result, he claims, was "spectacular" and "staggering," produced in 30 seconds at near-zero cost.
The experiment nobody wants to talk about
The barrister's claim is striking: that AI output on a redacted version of his own day-and-a-half work product matched "the level of a truly great KC." That's Queen's Counsel, for the uninitiated. The best of the best.
"How can any of us compete? We can't," he told The Spectator's Sean Thomas.
It's a dramatic assertion, and worth examining skeptically. The experiment he describes used a "real, recent and important case," but the details are thin. What kind of prompting was required? Did human lawyers review the output for accuracy? Were there hallucinations lurking in the citations? He doesn't say, though he waves away concerns about AI errors as "temporary bugs."
Reality check from the benchmarks
The barrister's confidence stands in sharp contrast to recent research. MIT Technology Review reported last week that leading AI models score just 37% on the most difficult legal problems according to ScaleAI's Professional Reasoning Benchmark. That's barely a third of possible points. The models "frequently made inaccurate legal judgments," and even when reaching correct conclusions, did so through "incomplete or opaque reasoning processes."
Afra Feyza Akyurek, the study's lead author, put it bluntly: AI tools "are not there to basically substitute your lawyer."
The Mercor AI Productivity Index tells a similar story. The best-performing model scored 77.9% on legal tasks, which sounds respectable until you consider what that means in practice. In fields where errors are costly, satisfying four out of five criteria may not be good enough.
Hallucinations are already happening in court
The timing of The Spectator piece is no coincidence. Just days earlier, a Scottish employment tribunal ruling became a cautionary tale about unchecked AI use in legal work.
Judge Alexander Kemp's 300-page judgment in the Sandie Peggie case contained at least two citations that appear to have been fabricated entirely. One quote, attributed to the Maya Forstater employment tribunal, simply doesn't exist in that judgment. "I know that judgment inside out," Forstater said. "Those words are not there."
The Judicial Office issued a certificate of correction and quietly swapped in a different quote, but refused to explain where the fictitious citations came from. J.K. Rowling called the situation "scandalous." The ruling is now being appealed.
UK courts did issue updated AI guidance in October warning that AI tools may "make up fictitious cases, citations or quotes." Judges are reminded they're "personally responsible for material which is produced in their name" and "must always read the underlying documents." Whether Judge Kemp followed that guidance is unclear. The Judicial Office isn't saying.
Where the money is actually going
The legal AI market is certainly real. Harvey, the most-hyped legal AI startup, reportedly hit $150 million in annual recurring revenue by November 2025, up from $50 million at the end of 2024. The company raised $150 million in October at an $8 billion valuation, its third fundraise of the year.
Major law firms are signing on. PwC, A&L Goodbody, Ashurst, and WongPartnership have all announced partnerships. Harvey handles document analysis, legal research, and drafting, claiming 80x speed improvements on certain tasks.
But there's a catch buried in Sacra's analysis of the company: Harvey "scrapped its proprietary vertical model after frontier reasoning models from Google, xAI, OpenAI, and Anthropic began outperforming Harvey's custom legal model on its own BigLaw Bench evaluation." The company now positions itself around workflow orchestration rather than model superiority. In other words, even the legal AI specialists are discovering their secret sauce isn't that secret.
What the barrister gets right (and wrong)
The anonymous barrister makes one observation that rings true: lawyers are arrogant. "Keir Starmer is a quintessential lawyer," he notes. "For them to admit they aren't so special, at all, and that they can be replaced by a free robot, is going to be torture."
He's probably right that the profession is underestimating the speed of change. A Harvard Law professor told Harvard Law Today in March that lawyers face "growing pressure to navigate this increasingly competitive ecosystem of providers." That includes Big Four accounting firms, legal tech companies, and consulting shops, all nibbling at work that once belonged exclusively to law firms.
But the barrister's timeline feels aggressive. "Maybe even the judges," he says of those AI will replace. Yet the Peggie case shows exactly why that's unlikely soon. When AI-generated errors appear in judicial rulings, they erode public trust in the legal system. Fixing those errors requires humans. So does figuring out who made them.
His advice to his niece? "Please do not destroy your life. Do not get into a lifetime of debt for a job that won't exist in ten years. Or less."
It's a strong stance. Whether it's correct depends heavily on which end of the legal profession you're looking at.
The Peggie ruling is currently under appeal. Harvey's next fundraising round will presumably be announced soon. The UK judiciary's AI guidance is scheduled for another review. The answers aren't here yet.
