In April 2023, as the world was still scrambling to come to terms with the long-term implications of generative AI, Goldman Sachs published a study predicting that the nascent technology could “expose the equivalent of 300 million full-time jobs to automation.” The report claimed that “knowledge workers,” a category generally assumed to include lawyers, would be most at risk.
More than two years on, though, the jobs of most attorneys will be safe for a while yet. According to a recent survey by legal insurer Embroker, 78% of US law firms were not using any AI tools at all as of year-end 2024.
While there are signals that the legal industry’s AI adoption curve is beginning to steepen, many parts of the industry still appear unwilling to embrace it.
The Current State of Play
Though the uptake of AI has been slow in law, it is starting to gather speed.
A global survey by Thomson Reuters found that the share of legal organizations actively integrating generative AI rose from 14% in 2024 to 26% in 2025, and that 45% of law firms either use it or “plan to make it central to their workflow within one year.”
However, this statistic tells us little about the practical extent of AI uptake.
Something in the Way
“Even the ‘bullish’ firms on AI are still hyper-cautious about it,” says Alex Shahrestani, founding partner of Promise Legal. Shahrestani, a practicing attorney with a degree in computer science whose firm also delivers tech solutions to other attorneys, believes the careful nature of the average lawyer is the key factor here. “There’s an obvious level of concern about AI in what is already a risk-averse industry. A lot of the conversation still revolves around, 'should I even use AI in the first place?' ”
However, Shahrestani also believes that this mindset cannot survive for much longer. “Using AI right now is an exercise in maintaining relevance. Attorneys need to be prepared for the day (that is certainly coming) when AI can do our jobs.”
So, what’s standing in the way?
Barriers to AI Adoption in Law
Confidentiality and Data Privacy
This is perhaps the biggest sticking point for law firms when it comes to AI adoption. Forty-one percent of respondents to Embroker’s 2024 survey of over 200 American lawyers reported concerns about data privacy related to the adoption of AI in practice.
While publicly available LLMs like ChatGPT do a reasonably impressive job on tasks like drafting and contract review, most attorneys don’t trust them with privileged client data. Larger law firms have gotten around this by developing in-house LLMs, as well as by subscribing to third-party technology providers like Harvey and Luminance. However, many small firms are unwilling to take the risk on such investments.
While smaller outfits can still use free AI models for drafting and review functions if they redact sensitive material, this adds significant friction to the process–which, at a certain point, effectively defeats its purpose. However, Alex Shahrestani thinks this problem has more to do with perception than reality.
“The huge irony here is that for the past 30 years, attorneys have been divulging the exact same information [that they would to an LLM] to all sorts of legal platform providers. On the back end, OpenAI is more or less the same as Westlaw in this respect. When you're talking about legal tech, and tech tools in general, the main protection is a contractual obligation of confidentiality, which LLMs also offer.
“Plus, with AI providers, there’s the added benefit that AI doesn’t actually read text. When you have that guy on the back end working at, say, Thomson Reuters, he’s looking at the plain text data that you've put into their software. Maybe you've used their automated drafting tool, and you've put in some client information; that is plainly visible in their database. In an AI database, without further effort, the actual representation is gibberish to the human eye.”
Accuracy and “Hallucinations”
In June 2023, attorneys for New York law firm Levidow, Levidow & Oberman were fined $5,000 for submitting a legal brief including six fictitious cases. The brief had, of course, been written with ChatGPT. Incidents like this loom large in the minds of conscientious lawyers. While drafting is one of the areas in which LLMs could potentially lead to massive time savings, it’s also the area in which careless use could cause the most damage.
Cayce Lynch, national managing partner at Tyson & Mendes, says “AI may give us the bare bones of a draft, but it is absolutely no substitute for our professional judgment and still requires substantial revision by a licensed attorney before it’s ready to go out the door.”
Of course, just because an LLM hallucinates, that doesn’t mean it’s useless.
“It's not a search engine. It's not going to retrieve information for you, it's going to predict what the answer might look like,” says Alex Shahrestani of LLMs. He believes that more sophisticated prompting techniques (such as the use of specific questions and additional context materials) is the solution to this problem.
Cost
One of the more interesting early narratives around AI adoption in legal firms was that it might break the stranglehold of Big Law on certain areas of the industry. However, emerging data suggests it’s doing exactly the opposite. According to a study published by the Federal Bar Association, American firms with 51 attorneys or more are using AI at roughly double the rate of firms with fewer lawyers than that. The price tag of firm-ready AI systems is the likely culprit here.
“Cost is significant,” Cayce Lynch notes. “We've made a strategic decision that we need to be investing in this now. Not every law firm has that option, and not every law firm will make that decision. I think there's still an element of fear in the legal industry as well around doing something new and unknown.”
So, it’s not always that law firms can’t afford to implement AI; for some, it’s that they don’t trust it enough to make the investment. Another issue is that smaller firms lack the understanding required to properly allocate AI resources, says Alex Shahrestani.
“Smaller firms are contemplating a monthly Westlaw subscription of $600 per user as their only option. But, if you hire a low code or no code expert to build out the system for you, it's pretty much dirt cheap. You're talking $10,000 all in for a robust system that does exactly what you want it to do. The last cost you have remaining is the overhead of running the server, which is also cheap. It's just not as expensive as people think.”
Integration Challenges
“Many AI tools fail adoption tests because they don’t integrate with existing workflows or demonstrate clear ROI. Firms will only adopt when a product reduces measurable time or risk without creating new work or liability,” says Alex Shahrestani.
AI is easier for firms to use when it’s woven into existing software. So, the more software-reliant a firm is already, the easier it will be for that firm to start using AI. Firms that are less tech-oriented will have a longer road to travel.
It’s also likely that firms simply haven’t considered the full range of functions they could potentially automate. Quality control and data privacy are huge concerns when it comes to mission-critical tasks (such as drafting), but they don’t factor into decisions around the automation of administrative functions nearly as much.
Take timekeeping, for example. Lawyers at Tyson & Mendes use Traced, an AI-driven tool that automatically tracks what they’re working on and drafts detailed billing entries for them. “It’s really improved our accuracy. It’s also provided a mental health benefit, because our lawyers aren’t having to track every minute of their day. It’s given folks relief,” says Cayce Lynch.
Regulations
Integration challenges are not solely the result of old habits. Some jurisdictions actively regulate old-school methods into continued existence.
In Ireland, for example, Law Society conveyancing guidelines require deeds and declarations to be executed using wet ink signatures only (it had the same provision in place in relation to contracts for sale until 2023).
Ireland is home to many of the world’s biggest tech offices (including, ironically, the EMEA headquarters of Docusign). When any of these mega-firms buy, sell, or lease property in Dublin, their solicitors have to prepare and issue documentation on paper, in much the same way lawyers have been doing for decades.
Ireland is not alone in this respect; plenty of other countries, and some US states, still have similar rules in certain practice areas. As of yet, there’s no agentic AI model that can print and bind property deeds before putting them in the mail. Where regulations like these exist, cutting-edge tech can only do so much.
AI and the Billable Hour
Earlier this year, Clio CEO Jack Newton predicted that “the billable hour model cannot survive” the AI generation.
Speaking to the England and Wales Law Society Gazette for an article published in June 2025, Newton described a “structural incompatibility” between AI-driven productivity gains and hourly billing – if an AI lets a lawyer accomplish in one hour what used to take five, their time-based invoice would shrink by 80%, despite their output being identical.
Newton is not alone in this view. In Thomson Reuters’ Generative AI in Professional Services Report 2025, 40% of law firm respondents believed that AI will lead to an increase in non-hourly billing methods.
Not So Fast
Lynch is less certain about the capacity of AI to drastically cut down on the amount of time lawyers spend on billable tasks, at least in its current form.
“We’re seeing the best time savings from AI around summarizing documents. When it comes to drafting, I wouldn't quite say we've found much of a productivity increase there yet.”
Additionally, the idea that law firm clients are all crying out for an end to hourly billing is mistaken, according to Lynch. “As unwieldy as the billable hour can be, it’s the way things always have been. So it’s still the comfort zone for a lot of clients. Some clients prefer hourly billing because it’s what they’ve always known, what they know how to measure. We’re in a client service business, and so we meet the needs and expectations of the client.”
This obviously won’t be the case in every practice area; many clients will much prefer a reliable flat fee than a variable hourly estimate. However, as long as there are at least some clients who demand hourly billing, the system will survive.
Can Law Firms Afford Not to Automate?
Despite the persistent issues around AI adoption, the river of change is only flowing in one direction. As lawyers get more familiar with the capabilities and limitations of AI models, they will naturally begin to seep into practice more and more.
The question of AI adoption in law is currently seen as one of risk. As adoption rates grow, though, this discussion will likely turn on its head. Firms that currently view AI usage as a gamble may come to realize that the risks of not adopting it may be even greater.
