Will AI Replace Lawyers? What the Data Says in 2026

Goldman Sachs published new workforce research in August 2025 estimating that AI-driven job displacement will affect about 2.5% of total US employment, with legal roles among the most exposed categories. 

A separate analysis of that same Goldman data, run specifically for the legal sector, puts attorney job risk at roughly 17%, a real number but nowhere near the "half the profession" fear that circulated a couple of years ago.

The shift mirrors what e-commerce did to retail. Online stores didn't eliminate brick-and-mortar shops, but they restructured which roles grew and which shrank. 

AI is doing something similar to legal work now, automating the repetitive parts of the job while leaving judgment, advocacy, and client relationships to people. This is where AI agents for legal teams can extend beyond simple task automation by coordinating multi-step workflows such as research, document review, intake, and matter management. 

So the short answer holds. But real movement is happening underneath it. The ABA's Task Force on Law and Artificial Intelligence closed out its two-year mandate in December 2025 with a clear conclusion. 

AI has stopped being a future consideration for the profession and has become core infrastructure for how lawyers practice, courts operate, and firms serve clients. 

This guide breaks down what AI still cannot do in legal practice, where the data actually shows it outperforming human lawyers, which legal jobs face the most disruption, and how firms are using it right now.

AI still falls short of human lawyers in six areas, all of them requiring judgment, adaptability, or personal accountability that current models cannot replicate.

1. Creative Legal Strategy

Legal problems sometimes need an argument nobody has made before. Defending a startup against a patent troll, a lawyer might find prior art buried in a foreign-language academic paper. That kind of cross-domain leap comes from intuition built over years of practice, not pattern matching against a training set.

2. Negotiation and Emotional Dynamics

In a high-stakes M&A negotiation, an experienced dealmaker can sense that the opposing general counsel is under board pressure to close before quarter-end. They adjust the term sheet to match. Reading that kind of pressure takes years of exposure to human behavior, not access to a legal database.

3. Emotional and Empathetic Client Representation

A lawyer representing a wrongful termination plaintiff might recognize that the client's real need is validation and a public acknowledgment of wrongdoing, not the largest possible settlement. The lawyer then negotiates for that. AI can optimize for maximum financial outcome. It cannot recognize when financial outcome isn't actually the client's goal.

4. Ethical and Contextual Judgment

The line between aggressive advocacy and misleading the court, or between hardball negotiation and coercion, is a judgment call. It takes ethical reasoning and professional responsibility training, not just rule matching. AI can flag potential issues. It cannot exercise the professional judgment needed to resolve them.

5. Adaptive Strategy in Dynamic Situations

Trial advocacy is live. When a witness says something unexpected on the stand, a human attorney adjusts the entire examination strategy in seconds. AI can generate a pre-planned set of questions. It cannot adapt in real time to a courtroom where the unexpected is routine.

6. Ethical Accountability

Lawyers are personally licensed and professionally liable for the advice they give. An AI cannot be sanctioned by a bar association, cannot be sued for malpractice, and cannot exercise the independent professional judgment required under ABA Model Rule 2.1. 

When an AI tool makes a mistake in a legal matter, a human attorney still bears the professional responsibility for it. That's why AI is correctly positioned as a tool a lawyer uses, not a replacement.

None of these are permanent walls. Emotional recognition is improving in multimodal models, legal reasoning is improving in frontier models, and the boundary keeps moving. But it hasn't moved far enough yet to change any of the six above.

The data is split, and more interesting than a simple "AI wins" or "humans win." On routine contract drafting, current AI tools already match or edge out human lawyers on reliability and are dramatically faster.

Some of these limits are technical and will narrow as models improve. Others, like licensure and personal accountability, aren't technical. No model upgrade changes who answers to the bar.

On judgment-heavy work, humans still win decisively, and AI's legal research tools carry a real accuracy problem of their own.

The clearest current evidence comes from a September 2025 benchmarking study, "Benchmarking Humans & AI in Contract Drafting," run by the independent research group LegalBenchmarks.ai. 

Researchers tested 13 AI tools against experienced in-house lawyers (10 years' experience on average) across 30 real-world contract drafting tasks, then had 72 practicing lawyers evaluate the outputs. Human lawyers produced a reliable first draft 56.7% of the time. 

The best-performing AI tool, Google's Gemini 2.5 Pro, hit 73.3%, edging out even the best individual human lawyer in the study at 70%. The speed gap was even starker. Lawyers took nearly 13 minutes per task. AI tools finished in seconds. 

The same study turned up a surprising result. In scenarios carrying real legal risk, specialized legal AI tools raised explicit risk warnings in 83% of outputs, and general-purpose tools like ChatGPT and Claude did so 55% of the time. 

In the same scenarios, human lawyers raised no risk warnings at all. In one task involving a penalty clause that was likely unenforceable under New York law, the AI tools flagged it, and the human lawyers didn't.

That doesn't mean humans lost across the board. The same research found lawyers clearly ahead on tasks requiring commercial judgment and multi-source context, correctly reading client intent and avoiding concessions that would hurt their side.

In one task requiring information pulled from a screenshot, a term sheet, and an email thread, only the human lawyer extracted the party details correctly. Every AI output got something wrong.

And on legal research specifically, AI still has a real reliability gap. A 2025 study published in the Journal of Empirical Legal Studies found that leading legal AI research tools, including Lexis+ AI and Westlaw's AI-Assisted Research, hallucinated or produced inaccurate legal citations 17-33% of the time, depending on the tool.

AI vs. Human Lawyers: Task-by-Task

The practical read is that AI is now useful as a first pass, and sometimes a sharper one than a rushed human review. But it hasn't closed the gap on judgment calls, and it introduces its own accuracy risk wherever it's doing research rather than drafting against a template. 

As one Fortune 500 general counsel put it in the LegalBenchmarks study, accuracy is only one factor in whether a tool gets used, and the lawyer stays responsible for the work product either way.

AI has a clear edge in four areas (speed, first-pass research, workflow efficiency, consistency) and fills a fifth gap lawyers rarely reach: access for people who can't afford one. None of these make AI a replacement, but together they explain why adoption keeps climbing.

1. Time and Cost Savings

The clearest advantage is raw speed. In the September 2025 LegalBenchmarks.ai study referenced above, human lawyers took nearly 13 minutes to complete a standard contract drafting task. AI tools did it in seconds. That reduction in manual work can lower the cost of routine legal tasks, particularly when AI handles high-volume work that would otherwise require significant lawyer hours. 

For example, Maruti Techlabs built a custom AI platform for an Am Law 200 firm with 1,000+ attorneys, handling document review, legal research, and drafting on complex 300 to 1,000+ page legal documents. Tasks that previously took the firm's attorneys 7 to 8 hours now take minutes, with over 95% accuracy in complex legal reasoning workflows.

2. Faster First-Pass Legal Research

AI research tools can surface relevant case law and statutes in seconds instead of hours, which is useful for a first pass. But that speed comes with a real accuracy cost, and it's worth stating plainly. 

3. Process and Workflow Efficiency

Beyond research and drafting, AI is well suited to the repetitive backbone of legal work: 

• Calendaring deadlines
• Generating compliance checklists
• Flagging conflicts and organizing case files

This is lower-stakes automation, but it adds up across a caseload and frees attorney time for judgment work.

4. Expanding Access to Basic Legal Understanding

This is AI's most significant benefit outside the billable-hour economy. An estimated 75% of the roughly 20 million civil cases filed in US state courts each year involve at least one party without a lawyer. 

Stanford's Legal Design Lab is actively piloting AI tools with legal aid organizations and courts specifically to help unrepresented people understand notices and navigate cases they'd otherwise face alone. AI isn't replacing lawyers for people who have one. It's giving basic legal understanding to people who never had access to a lawyer in the first place.

5. Consistency and Reduced Variability

Human review quality varies by reviewer, fatigue, and time of day. AI applies the same rule to every document, every time. 

The September 2025 benchmarking study caught this directly. In a task requiring a contract to state a 10% penalty clause, every AI tool reproduced the figure correctly. One human lawyer in the study wrote "9%" by mistake. In compliance-heavy work, such as financial services, healthcare, and real estate, removing that kind of slip is worth a lot.

Risk isn't evenly spread across the legal profession. It follows a clear pattern: the more a role depends on processing information the same way every time, the more exposed it is. The more it depends on judgment, advocacy, or relationships, the safer it is.

Higher-risk roles

• Document review specialists. This is where AI displacement is most real and most current, especially in large e-discovery matters. AI-assisted review tools now handle first-pass classification and issue-spotting across hundreds of thousands of documents, a task that used to require teams of contract attorneys working around the clock.
• Paralegals focused on document processing. Where a paralegal's core function is drafting standard documents, filling forms, and checking research citations, AI tools now perform those specific tasks in minutes. The role isn't disappearing, but it's being restructured around the tasks AI can't do, like managing exceptions and unusual fact patterns.
   

Medium-risk roles

• Contract associates. Junior associates who used to do first-draft contract review and redlining are seeing that work shift. AI does the first pass now, and the associate's job is shifting from producing the draft to supervising and correcting it. Billing structures built around junior associate hours are under real pressure here.
• Legal researchers. AI handles citation pulling and case law summarization well. A researcher's value now lies in knowing what to search for and judging whether the results are right, not in running the search. 
   

Lower-risk roles

• Trial attorneys. Courtroom advocacy remains entirely human, in every jurisdiction. No AI system can currently cross-examine a witness, read a jury, or respond to a judge's question it didn't anticipate.
• Specialized counsel in M&A, restructuring, and complex litigation strategy. The higher the stakes, the more the work depends on relationship management, strategic judgment, and creative deal-making, all of which stay firmly human.
   

For scale, the World Economic Forum's Future of Jobs Report 2025 projects 92 million jobs displaced globally by 2030 against 170 million created, with 39% of workers' current skills expected to be transformed or made obsolete within five years. 

This is an economy-wide figure, not a legal-specific one, but it captures what's happening inside law too: less job loss than skill and task restructuring.

A&O Shearman + Harvey

A&O Shearman, the nearly 4,000-lawyer firm formed by the 2024 Allen & Overy/Shearman & Sterling merger, was the first Magic Circle firm to deploy generative AI firm-wide, trialing Harvey with 3,500 lawyers as early as 2022. 

In 2023, the firm launched ContractMatrix, an AI contract tool built with Harvey and Microsoft. In April 2025, it added agentic AI agents for antitrust filings, cybersecurity, fund formation, and loan review.

The firm has not reported any headcount reductions. Its own lawyers describe the shift as freeing time for higher-value strategic work rather than replacing roles.

Thomson Reuters CoCounsel at Polsinelli

Am Law 100 firm Polsinelli rolled out CoCounsel firm-wide to its roughly 1,200 attorneys, with 70% of associates and 50% of shareholders actively using the tool for document discovery, review, and analysis. 

Polsinelli's chief practice strategy officer described it as a firm-wide innovation initiative rather than a headcount-reduction move, framed around handling more client work with the same team.

The pattern

Across both examples, the story is the same. AI is expanding what existing legal teams can handle, not shrinking legal employment, at least so far. The more useful question for 2026 isn't whether AI replaces lawyers. It's whether law schools and junior attorneys are adjusting their skills toward the work AI can't do.

No. But the legal profession in five years will not look like the one it is today.

AI now handles repetitive, pattern-based parts of legal work well. It drafts a first-pass contract in seconds, and in the LegalBenchmarks.ai study, it edged out human lawyers on draft reliability. It flags risk in standard clauses at rates that beat rushed human review. It processes legal research faster than any associate could, even with real, documented accuracy gaps that still require a human to check every citation.

What it still can't do is everything this guide's first section covers. It can't build a courtroom relationship with a judge, read a jury, or shift examination strategy mid-trial. It can't sense that a client wants acknowledgment more than money. 

It can't be sanctioned by a bar association or held personally accountable when it gets something wrong. Some of these gaps will narrow as models improve. Others won't, because they aren't technical. A license, a malpractice policy, and a duty to the court belong to a person.

The result is a collaborative split, not a replacement. AI takes the volume work. Lawyers keep the judgment work. A&O Shearman and Polsinelli both show what that looks like in practice: firms expanding what their existing teams can handle, not shrinking their teams.

The lawyers best prepared for 2030 will know where AI adds value and where human judgment still matters. They know which tasks AI does better, and which tasks still require irreplaceable human judgment. That combination, AI efficiency paired with human judgment, is what legal practice looks like from here.

1) Will AI replace lawyers?

No, not as a profession. AI is already automating specific legal tasks and reshaping some entry-level roles, but it can't be licensed, held professionally accountable, or exercise the independent judgment a lawyer's job legally requires. The lawyers most exposed are those doing high-volume, low-judgment work like document review and standard contract drafting. Those least exposed do complex advisory, advocacy, and strategic work.

2) Can AI practice law?

No. In the US, practicing law requires a license, and AI cannot hold one. A few states, such as Arizona and Utah, allow approved non-lawyer providers under strict regulation, but a licensed human remains responsible. The ABA's 2024 Formal Opinion 512 makes clear that attorneys remain professionally responsible for all AI-generated work product, regardless of how it was produced.

3) Which type of lawyer is most at risk from AI?

Document review specialists and paralegals doing routine research and drafting face the most near-term disruption. Junior associates at large firms are seeing their role shift from producing first drafts to supervising and correcting AI output. 

Trial attorneys, senior advisors, and specialists handling novel legal questions face the least disruption, since that work depends on judgment AI doesn't have.

4) Is AI better than lawyers at legal research?

AI can be faster than lawyers at searching large volumes of legal information, identifying relevant cases, summarizing documents, and surfacing patterns. However, it is not consistently better at legal research overall. 

Lawyers still need to assess whether sources are authoritative, interpret legal precedent in context, identify jurisdiction-specific nuances, and verify AI-generated findings. AI works best as a research assistant that accelerates information discovery, while legal professionals remain responsible for judgment and verification.

5) What legal tasks can AI already do in 2026?

AI currently handles contract review and redlining, legal research with citation support, e-discovery document classification, litigation analytics, deposition summarization, compliance checklist generation, and first-draft document creation from templates. 

It still can't represent a client in court, give licensed legal advice, exercise independent professional judgment, or navigate a novel legal question without human direction.

6) Will studying law still be worth it in 2026?

Yes, but the skills that matter are shifting. Law students should prioritize learning to supervise and evaluate AI-generated work, build expertise in areas AI can't automate (litigation strategy, transactional judgment, policy advocacy), and develop the client relationship and business development skills AI can't replicate. 

Employment of lawyers is projected to grow 4% from 2024 to 2034, about as fast as the average for all US occupations, according to the Bureau of Labor Statistics' most recent projections. That's slower growth than some other fields, but it's growth, not decline.

The client is a leading national law firm with over 500 attorneys and government professionals, serving regional, national, and international clients, including more than 40% of Fortune 500 companies. The firm sat at the center of syndicated loan transactions, where every deal involved a Credit Approval, a Term Sheet, and a Credit Agreement that all had to be reviewed and kept aligned.

The documents were long, dense, and unstructured, which slowed extraction and interpretation. Inconsistencies between deal documents were easy to miss. Clause-by-clause validation held up deal timelines, and every missed discrepancy carried legal, financial, and compliance risk. The firm also lacked a standard framework, such as an LSTA-based comparison, to assess deviations consistently across deals.

Maruti Techlabs built an AI-powered syndicated loan platform that automates document processing, validation, and risk analysis in one workflow. The platform ingests multiple deal documents asynchronously and converts unstructured PDFs into structured, machine-readable data. It then compares clauses against LSTA-based standards to classify deviations and validates consistency across the Credit Approval, Term Sheet, and Credit Agreement.

The platform extracts 32 critical data points for structured analysis and drafts commitment letters automatically from validated inputs. Attorneys can also chat with the documents in natural language to get quick answers without searching manually. The result is a system built around how lawyers actually review a deal, with the AI handling repetitive comparison work and attorneys keeping the judgment calls.

The impact:

• 60-70% reduction in manual review effort
• 3x faster deal processing
• 95%+ improvement in accuracy
• 40% reduction in risk exposure
• 2x increase in team productivity
• 50% faster access to key deal information