In 2019 a steel cart rolled down the aisle of an Avianca flight from El Salvador to New York and struck Roberto Mata on the knee. Mata retained two private harm attorneys to sue the airline, leading to a case that made authorized precedent—however not for a motive Mata or his attorneys would have hoped for. As an alternative, the 2 attorneys achieved notoriety in 2023 when the decide within the case fined the pair $5,000 for citing six fictitious instances, invented by synthetic intelligence software program, in a quick to the court docket.
The punishment was the primary of its variety and was broadly reported in authorized circles as a cautionary story for attorneys concerning the threat of counting on AI. That cautionary story has not been heeded. Within the two years because the incident, there have been no fewer than 139 instances of authorized briefs containing AI-invented citations—together with over 30 this Could alone. Whereas many of those got here from self-represented litigants, in addition they embrace briefs filed by outstanding nationwide regulation corporations. And, in a single occasion, a decide struck knowledgeable testimony submitted by a knowledge scientist from AI big Anthropic—on the grounds it contained a “hallucination” created by Anthropic’s personal software program.
These tales of AI gone mistaken within the courtroom make for a energetic supply of chatter round regulation agency watercoolers. But these instances are additionally outliers. In actuality, the overwhelming majority of attorneys utilizing AI are doing so in accountable style—together with by verifying that AI-produced citations truly try earlier than placing them earlier than a decide.
Many are additionally delighted by how helpful it’s. Attorneys say AI excels at analysis and doc evaluate and that, extra just lately, they’ve discovered it to be able to tougher duties like drafting authorized briefs. And all of this progress has massive implications for each the long run trajectory of authorized careers and for the enterprise of regulation.
A rising suite of AI instruments
Most attorneys got here to understand the transformative energy of AI similtaneously everybody else: November 2022. That was when OpenAI launched the preliminary model of ChatGPT to most people, and, unsurprisingly, it was about six months later that the primary wave of pretend citations started displaying up in courtrooms.
In reality, the thought of utilizing AI to tell authorized analysis has been round for over a decade. Pioneers within the discipline embrace the likes of Ironclad and DISCO, which respectively provide automated methods to create and handle contracts and to expedite the invention course of.
The distinction immediately is that, as in different industries, the previous three years have seen the authorized occupation purchase and deploy AI tech at a broader and sooner tempo than ever earlier than—particularly as generative AI has grow to be extra simply accessible and extra highly effective.
In February, the Federal Bar Affiliation revealed a survey of over 2,800 authorized professionals: 21% stated they used AI of their apply in 2024, and the quantity was considerably greater—39%—for attorneys at massive regulation corporations. A separate lawyer survey revealed in April by the authorized transcription service Rev discovered a lot greater general adoption, with 48% of respondents saying they used AI for analysis.
Unsurprisingly, the variety of AI choices aimed on the authorized occupation is rising. There are providers like Harvey, which is constructed on prime of AI fashions from corporations together with OpenAI, Anthropic, and Google and makes a speciality of drafting and analysis; and Vancouver, British Columbia–based mostly Clio, which gives apply administration instruments. Clio raised $900 million in 2024, valuing the corporate at $3 billion; in late June, Harvey raised $300 million in a funding spherical that values the corporate at $5 billion.
For now, there is no such thing as a clear knowledge on how broadly the attorneys deploying AI of their jobs are utilizing these specialised instruments, or if they’re as an alternative counting on fashionable LLMs created by the likes of OpenAI and Google, which have all ingested reams of common-law statutes and choices.
In accordance with Mark Lemley, a Stanford mental property professor properly versed in authorized tech, most regulation college students and junior attorneys are doubtless utilizing normal providers like ChatGPT, since these are probably the most accessible.
In massive corporations, a well-liked AI device is CoCounsel, which was developed by the authorized startup Casetext. The info big Thomson Reuters bought Casetext in 2023 and now gives CoCounsel as a part of Westlaw, a well-liked however costly service broadly utilized by so-called Massive Legislation corporations.
Daryl Landy, a litigator with the nationwide agency Morgan Lewis, says he likes CoCounsel partially as a result of it’s designed to take account of the authorized occupation’s particular obligations concerning shopper confidentiality and knowledge safety. In apply, which means AI prompts created by Landy and his colleagues don’t grow to be a part of a coaching set that may inform the analysis of individuals exterior his agency—an vital difficulty in a occupation by which practitioners usually have entry to extremely delicate details about their purchasers.
Altering skilled norms
For many years, most regulation corporations have match neatly into the “finders, minders, and grinders” mannequin. The phrase describes a pyramid construction the place a handful of rainmakers on the prime usher in purchasers and hand over the majority of the work to a second tier of senior attorneys, who in flip oversee a bigger pool of associates.
The place does an AI agent match into the pyramid? It’s been apparent for a while that AI is an effective match to displace massive swaths of mundane “grinder” work—and certainly that’s occurring with duties like doc evaluate, contract drafting, and fundamental analysis. It is a good factor in lots of respects. AI is already eliminating a few of the drudgery of authorized work, a lot of which has historically come about in pursuit of billable hours—a infamous side of the occupation that always leaves purchasers questioning what number of hours on an bill signify busywork reasonably than indispensable duties.
On this gentle, it’s onerous to argue with a know-how that vastly reduces time spent on tedious duties, reduces prices, and guarantees to liberate attorneys for deeper and extra strategic issues. Nonetheless, there are drawbacks, most notably in relation to coaching the subsequent era of attorneys.
“A technique you get good authorized instincts for logical arguments is by getting the reps in—attempting issues, and doing them time and again,” says Lemley. He worries that wider adoption of AI means younger attorneys is probably not gaining a few of the analytic expertise which have historically been acquired by means of intensive studying of case regulation and statutes.
Lemley additionally observes that AI is more and more taking over duties like writing briefs and crafting arguments, which many attorneys view because the enjoyable a part of the job, and which generally fall to extra senior attorneys.
All of this raises the query of not simply how attorneys will spend their time sooner or later, but additionally how they’ll cost for his or her providers. For years, purchasers have fought—usually with restricted success—to introduce various payment preparations with a purpose to escape paying by the hour for junior associates to go down authorized rabbit holes.
Sooner or later, Lemley says, the state of affairs could also be reversed as purchasers—realizing that AI can carry out many authorized duties in minutes not hours—ask to be billed for mounted blocks of time, whereas regulation corporations suggest flat charges and different preparations.
The authorized occupation has lengthy been adept at guarding its enterprise mannequin, so such modifications are doubtless not on the rapid horizon. However for now, it’s turning into clear that AI is already offering attorneys with one thing treasured: extra time.
In accordance with the Thomson Reuters survey, the widespread adoption of AI will doubtless liberate 4 hours per week for the standard lawyer, or 200 hours a yr—hours that may very well be used to increase their shopper base, develop extra expertise or, after all, stack billable hours.
A rule of regulation for AI attorneys
Pablo Arredondo, who created the favored CoCounsel AI device and is now a vice chairman at Thomson Reuters, continuously runs seminars for judges concerning the intersection of regulation and synthetic intelligence.
In the middle of his work, he usually displays on a virtually century-old phrase that seems within the very first part of the Federal Guidelines of Civil Process—particularly, that judges ought to encourage the “simply, speedy, and cheap” decision of each motion.
“These three phrases usually are not how anybody would describe U.S. litigation,” Arredondo says dryly.
This might change if courts undertake AI instruments in a widespread and efficient style. And, certainly, such adoption is already starting, together with in Massachusetts the place the court docket system is adopting automated instruments to assist tenants dealing with eviction navigate complicated authorized varieties.
“To the extent this know-how can be utilized for higher outcomes, there’s a ethical crucial to make use of it,” says Arredondo. Nonetheless, he cautions that any deployment of AI by each courts and attorneys should happen with the strictest of oversight.
Arredondo is just not alone in that sentiment because the authorized occupation, like so many different industries, tries to stability the effectivity positive aspects supplied by AI with security. In 2023, reflecting on how the know-how is poised to disrupt the work of judges, none aside from Chief Justice John Roberts wrote, “Any use of AI requires warning and humility.”
