CtrlK
AI in law refers to the application of artificial intelligence technologies to legal research, document analysis, contract management, litigation support, and access to justice. The legal profession has adopted AI tools at an accelerating pace since 2023, driven by the capabilities of large language models (LLMs) and the entry of well-funded startups alongside established legal technology providers. At the same time, high-profile incidents of AI-generated fabricated case citations have highlighted the risks of uncritical reliance on these tools, prompting courts and bar associations to issue new guidance on responsible AI use in legal practice.
AI is being applied across nearly every stage of legal work. The table below outlines the major application areas, the leading tools in each category, and how they function.
| Application area | Leading tools | Description |
|---|---|---|
| Legal research | CoCounsel Legal (Thomson Reuters), Lexis+ AI (LexisNexis), Harvey AI | AI-powered search and analysis of case law, statutes, and legal commentary; natural language queries rather than Boolean searches |
| Contract analysis | Ironclad Jurist, Luminance, Kira Systems | AI review of contracts to identify risks, compliance gaps, key clauses, and obligations; automated redlining and playbook generation |
| Document review | Relativity aiR, Everlaw AI Assistant, DISCO AI | AI-assisted review of large document sets in litigation and investigations; prioritization of relevant documents, privilege detection |
| Due diligence | Luminance, Kira Systems, Diligence by DFIN | AI analysis of transaction documents in M&A, identifying risks, obligations, and anomalies across thousands of documents |
| Legal drafting | Harvey AI, CoCounsel Legal, Lexis+ AI, Spellbook | AI generation and editing of legal memoranda, briefs, contracts, and correspondence |
| Prediction and analytics | Lex Machina, Pre/Dicta, Gavelytics | Analysis of judge behavior, case outcomes, and litigation trends to inform legal strategy |
| Access to justice | LawDroid, A2J Author, DoNotPay | AI chatbots that provide self-represented litigants with legal information, procedural guidance, and help filling out court forms |
| Compliance AI | Streamline AI, Casepoint, Onit | AI tools that monitor regulatory changes, track compliance obligations, and automate reporting |
Legal AI tool pricing varies widely based on firm size, feature set, and deployment model. The following table summarizes publicly available pricing information as of early 2026:
| Tool | Provider | Target market | Approximate pricing | Key differentiator |
|---|---|---|---|---|
| Harvey AI | Harvey | Am Law 100 firms | Enterprise pricing (custom) | Deep law firm partnerships (A&O Shearman, Paul Weiss); $8B valuation |
| CoCounsel Legal | Thomson Reuters | All firm sizes | Included with premium Westlaw subscriptions; enterprise tiers available | Grounded in Westlaw proprietary legal content; agentic AI features |
| Lexis+ AI | LexisNexis | All firm sizes | Included with premium Lexis subscriptions | Integrated with LexisNexis legal database; linked citations |
| Spellbook | Rally Legal / Spellbook | Solo to mid-size firms | ~$179-199/seat/month (annual plans) | Focused on contract drafting; integrates with Microsoft Word |
| LegalOn | LegalOn Technologies | All firm sizes | Tiered pricing; free tier available | Best overall for contract review (industry rankings) |
| Luminance | Luminance | Mid-size to large firms | Enterprise pricing (custom) | Specialized for M&A due diligence and contract analysis |
| Clio Duo | Clio | Solo to mid-size firms | Included with Clio Suite subscription | Integrated practice management with AI legal research |
| EvenUp | EvenUp | Personal injury firms | Per-demand pricing | AI-generated demand letters for personal injury claims |
Legal-specific AI tools offer distinct advantages over general-purpose LLMs, including better case law knowledge, data privacy protection, and more accurate, comprehensive results. A Stanford HAI study found that legal AI models hallucinate in approximately 1 out of 6 or more benchmarking queries, underscoring the importance of specialized tools with verification features [17].
Harvey AI is a legal AI platform built on OpenAI's GPT-4 and subsequent models. The company has become the most prominent AI startup focused specifically on the legal profession.
Harvey's earliest and most significant partnership was with Allen & Overy (now A&O Shearman, following its 2024 merger with Shearman & Sterling), one of the world's largest law firms. Allen & Overy began deploying Harvey across its global practice in 2023, making it the first major law firm to roll out a generative AI tool firm-wide. The partnership gave Harvey a reference account that attracted additional clients from the AmLaw 100 [1].
In 2025, Harvey raised $760 million in total funding across multiple rounds, including a $160 million round in December 2025 that valued the company at $8 billion. The company's competitive position rests on its reference accounts at top firms (including A&O Shearman and Paul, Weiss), its investors, and a top-down adoption strategy that targets senior partners rather than individual associates [2].
In June 2025, Harvey announced a partnership with Paul, Weiss, Rifkind, Wharton & Garrison, which became the first law firm to use Harvey's "Workflow Builder" to create custom automated workflows for tasks like motions to dismiss and motions for summary judgment. Harvey also partnered with LexisNexis, gaining access to one of the two major proprietary U.S. legal databases [3].
Thomson Reuters, the parent company of Westlaw, entered the legal AI market through its $650 million acquisition of Casetext in August 2023. Casetext had developed CoCounsel, one of the first GPT-4-powered legal AI assistants.
In August 2025, Thomson Reuters launched CoCounsel Legal, a new platform combining agentic AI workflows with deep research capabilities grounded in Westlaw's proprietary legal content. The platform can automate legal tasks including document search and review, document and deposition summarization, case timeline preparation, deposition question drafting, trial strategy analysis, contract analysis, and brief drafting [4].
CoCounsel is used by over one million professionals in more than 107 countries. In April 2025, Thomson Reuters secured the federal judiciary contract for Westlaw Precision with CoCounsel, providing access to over 25,000 federal legal professionals [5].
Thomson Reuters announced agentic AI features for CoCounsel Legal launching in early 2026, including autonomous document review and "Deep Research" capabilities that can conduct multi-step legal research with minimal human direction. This reflects a broader industry trend toward AI agents that can execute complex legal tasks rather than simply answering individual queries [6].
Ironclad is an enterprise contract lifecycle management (CLM) platform recognized as a Leader in the 2025 Gartner Magic Quadrant for CLM. Its AI suite, branded "Jurist," includes specialized agents for different contract tasks:
| Agent | Function |
|---|---|
| Review Agent | Analyzes contracts against playbooks, identifies risks and deviations |
| Drafting Agent | Generates initial contract drafts from templates and negotiation history |
| Editing Agent | Produces first-pass redlines and suggests edits based on company playbooks |
| Research Agent | Conducts legal research with Bluebook citations across 60+ verified databases |
| Manager Agent | Orchestrates the other agents, managing multi-step contract workflows |
Ironclad's approach of using multiple specialized agents coordinated by a manager agent reflects the broader trend toward agentic AI architectures in legal technology [7].
Modern AI contract review tools offer capabilities that go far beyond simple keyword search:
| Capability | Description | Benefit |
|---|---|---|
| Playbook enforcement | AI compares contract terms against firm or company playbooks and flags deviations | Ensures consistency and reduces risk |
| Automated redlining | AI generates tracked-changes markup showing recommended edits | Saves hours per contract in negotiation |
| Obligation extraction | AI identifies and catalogs obligations, deadlines, and renewal terms | Reduces risk of missed deadlines |
| Risk scoring | AI assigns risk scores to clauses based on deviation from preferred terms | Prioritizes human review on highest-risk items |
| Clause library matching | AI compares contract clauses against a library of approved alternatives | Speeds drafting and negotiation |
| Multi-language support | AI reviews contracts in multiple languages with consistent analysis | Enables global contract management |
AI contract management can reduce legal review time by up to 80%, according to industry benchmarks [18].
Legal AI adoption has accelerated dramatically but remains uneven across firm sizes:
| Metric | Value | Source |
|---|---|---|
| Legal professionals personally using generative AI at work (2025) | 31% (up from 27% in 2024) | Clio Legal Trends Report |
| Legal professionals using at least one AI tool (2025) | 92% | Industry surveys |
| Firm-wide AI adoption (firms with 51+ lawyers) | 39% | AllRize report |
| Firm-wide AI adoption (firms with 50 or fewer lawyers) | ~20% | AllRize report |
| Lawyers who say AI is already "essential" to practice | 55% | AllAboutAI |
The gap between individual use and firm-wide adoption reflects the challenges of deploying AI at an organizational level: concerns about data security, client confidentiality, ethical obligations, and the need for governance frameworks that many smaller firms lack [19].
The most widely reported incident of AI misuse in legal practice occurred in Mata v. Avianca, Inc., a personal injury case filed in the U.S. District Court for the Southern District of New York. Roberto Mata sued Avianca airlines in February 2022, alleging injury from a metal serving cart during a flight.
Mata's attorneys, Steven A. Schwartz and Peter LoDuca of Levidow, Levidow & Oberman, used ChatGPT to conduct legal research for a brief opposing Avianca's motion to dismiss. The brief cited at least six fabricated cases, including fictitious decisions titled Martinez v. Delta Air Lines, Zicherman v. Korean Air Lines, and Varghese v. China Southern Airlines. The citations included fake quotations and internal citations that appeared plausible but corresponded to no actual court decisions [8].
The fabrications were discovered when Avianca's attorneys told Judge P. Kevin Castel that they could not locate the cited cases in legal databases. Schwartz testified that he had been "operating under the false perception that [ChatGPT] could not possibly be fabricating cases on its own" and said he had even asked ChatGPT to confirm the cases were real, which it did (falsely).
On June 22, 2023, Judge Castel sanctioned Schwartz, LoDuca, and their law firm, imposing a $5,000 fine. In his ruling, the judge emphasized the "gatekeeping role" that attorneys must play "to ensure the accuracy of their filings" and noted that the lawyers' conduct was "unprecedented" in its nature [9].
The case became a watershed moment for the legal profession's engagement with AI, prompting courts across the country to issue new rules and guidance on AI use in legal filings.
The Mata v. Avianca case was not an isolated event. AI-generated hallucinations in court filings have accelerated dramatically: documented cases rose from 120 total between April 2023 and May 2025 to 660 by December 2025, with new incidents occurring at a rate of four or five per day [17].
These incidents consistently involve attorneys using general-purpose AI chatbots (particularly ChatGPT) for legal research without verifying the results against authoritative legal databases. The phenomenon of AI-generated fabricated citations is a consequence of hallucination, a known limitation of LLMs in which models generate plausible-sounding but fictitious information.
A Stanford HAI study published in 2025 benchmarked legal AI hallucination rates and found that even purpose-built legal AI tools hallucinate in approximately 1 out of 6 queries, though rates vary significantly across platforms and query types. General-purpose LLMs produce hallucinated legal content at substantially higher rates [17].
Beyond attorney use of AI, courts themselves are beginning to adopt AI tools for administrative and judicial functions:
| Court AI application | Description | Examples |
|---|---|---|
| Case management | AI assists with docket management, scheduling, and case assignment | Several federal courts piloting AI-assisted scheduling |
| Sentencing analysis | AI tools that analyze sentencing patterns for consistency (advisory, not determinative) | Used in some state courts to identify potential disparities |
| Form completion assistance | AI helps self-represented litigants complete court forms | Multiple state courts deploying AI chatbots for form guidance |
| Transcript processing | AI transcription and summarization of court proceedings | Reduces backlog and improves accessibility |
| Legal research for judges | AI-assisted legal research for judicial chambers | Thomson Reuters' federal judiciary contract provides CoCounsel to 25,000+ federal legal professionals [5] |
The ABA published guidelines in February 2025 for responsible AI use by state and federal courts, addressing judicial use of AI for case management, scheduling, and administrative tasks. The Ohio Supreme Court has published an extensive AI resource library for courts, and multiple states have issued guidance on judicial AI use [13][20].
One of the most promising and least controversial applications of AI in law is improving access to justice for people who cannot afford legal representation.
In the United States, approximately 92% of low-income Americans cannot obtain legal help for their civil legal problems. Courts are filled with self-represented litigants who must navigate complex procedures without legal training. AI tools have the potential to narrow this gap by providing basic legal information, helping fill out court forms, and guiding people through procedural requirements.
Generative AI chatbots are being deployed by courts and legal aid organizations to help self-represented litigants. These chatbots provide legal information (not legal advice, an important distinction), procedural guidance, and help completing forms tailored to specific court systems.
LawDroid launched LawAnswers AI in September 2025, a nationwide platform designed to provide free legal information to the public. Thomson Reuters has an "AI for Justice" program, and Everlaw offers "Everlaw for Good," both aimed at providing advanced legal technology to public interest organizations at reduced or no cost [10].
The American Bar Association (ABA) has documented more than 100 AI use cases in legal aid settings. However, a significant concern remains pricing accessibility: the most reliable AI legal tools are expensive, and the ABA has warned that high subscription costs risk "widening rather than narrowing the justice gap" if legal aid organizations are priced out [11].
Following Mata v. Avianca, courts across the United States rapidly adopted rules governing AI use in legal filings. These range from disclosure requirements (attorneys must state whether AI was used in preparing a filing) to outright prohibitions on submitting AI-generated content without human verification.
The ABA issued Formal Opinion 512 in July 2024, addressing lawyers' ethical obligations when using generative AI. The opinion clarified that existing rules of professional conduct (including duties of competence, diligence, and candor toward the tribunal) apply fully to AI-assisted work. Lawyers must understand the capabilities and limitations of AI tools, verify AI-generated output, and maintain client confidentiality when using AI systems [12].
In December 2025, the ABA Task Force on AI reported that AI had "moved from experiment to infrastructure" for the legal profession, with the focus shifting from whether to use AI to how to govern, supervise, and integrate it responsibly.
Over 30 states have released AI-specific guidance for attorneys. Notable examples include:
| Jurisdiction | Key requirements |
|---|---|
| Pennsylvania | Mandatory disclosure of AI use in all court submissions |
| New York | At least two annual CLE credits in practical AI competency required (deadline Q3 2025) |
| California | Proposed rules requiring disclosure and attorney certification of AI-assisted filings |
| Florida | Standing order requiring attorneys to certify that AI-generated content has been verified |
| Texas | Local court rules requiring disclosure of AI use in briefs and motions |
The ABA also published guidelines in February 2025 for responsible AI use by state and federal courts, addressing judicial use of AI for case management, scheduling, and administrative tasks [13].
The rise of AI legal tools has revived debates about the unauthorized practice of law (UPL). If an AI chatbot provides legal information that is specific enough to constitute legal advice, it could be considered practicing law without a license. Bar associations are grappling with how to regulate AI tools that operate in the gray area between legal information and legal advice, particularly when these tools are used directly by consumers rather than by licensed attorneys [14].
The two dominant legal research platforms, Thomson Reuters (Westlaw/CoCounsel) and LexisNexis (Lexis+ AI), have each integrated AI into their products, but with different approaches:
| Feature | CoCounsel Legal (Thomson Reuters) | Lexis+ AI (LexisNexis) |
|---|---|---|
| Underlying legal database | Westlaw | LexisNexis |
| AI model | Custom models with GPT integration | Custom models with multiple LLM providers |
| Agentic capabilities | Deep Research, autonomous document review (launching 2026) | Multi-step research workflows |
| Citation verification | Citations linked to Westlaw sources | Citations linked to Lexis sources |
| Federal judiciary access | Exclusive federal judiciary contract (25,000+ professionals) | N/A |
| Users | 1M+ professionals in 107+ countries | Integrated into existing Lexis subscriber base |
| Harvey AI integration | No | Yes (Harvey has LexisNexis partnership) |
| Pricing | Premium Westlaw subscription tiers | Premium Lexis subscription tiers |
Harvey AI represents a third approach: a specialized legal AI platform that partners with law firms directly (A&O Shearman, Paul Weiss) and integrates with LexisNexis for legal data access. Harvey's Workflow Builder enables firms to create custom automated workflows for specific legal tasks [3].
The legal AI market in 2025-2026 is organized around three main categories of providers:
| Category | Key players | Strategy |
|---|---|---|
| Integrated legal research platforms | Thomson Reuters (CoCounsel/Westlaw), LexisNexis (Lexis+ AI), Clio | AI integrated into existing legal research databases with proprietary legal content |
| Specialized legal AI startups | Harvey AI, Spellbook, EvenUp, Clearbrief | Purpose-built AI tools for specific legal workflows, often partnering with established platforms |
| General-purpose AI providers | OpenAI (ChatGPT), Anthropic (Claude), Google (Gemini) | General LLMs used for legal tasks, either directly or as the underlying models for specialized tools |
The emerging trend is consolidation: Thomson Reuters and LexisNexis are building AI directly into their research platforms, Harvey is forming partnerships with data providers (LexisNexis) and law firms, and smaller specialty vendors face pressure to integrate with larger platforms or risk being acquired [15].
A notable development in early 2026 was the growing distinction between "vertical" legal AI (platforms like CoCounsel and Harvey that are purpose-built for legal work and grounded in authoritative legal databases) and "horizontal" AI (general-purpose models used for legal tasks). Legal professionals increasingly recognize that general-purpose chatbots pose higher risks of hallucination and error because they lack the guardrails and source verification built into dedicated legal AI platforms [16].
AI is beginning to reshape legal billing models. Industry analysts forecast that alternative fee arrangements (AFAs) will continue to grow as AI reduces the hours required for tasks traditionally billed by the hour. If AI can complete in minutes what previously took associates hours, the billable hour model faces pressure:
The legal profession's adoption of AI accelerated dramatically in 2025. Several trends define the current moment:
Harvey AI's $8 billion valuation and Thomson Reuters' launch of CoCounsel Legal with agentic capabilities signal that legal AI has reached enterprise scale. Fifty-five percent of lawyers now say AI is "essential" to their practice, and 92% report using at least one AI tool.
The hallucination problem remains the profession's primary concern. With documented fabricated citation incidents rising to 660 by December 2025, the gap between general-purpose LLMs and specialized legal AI tools has become a central issue. Stanford HAI's finding that even specialized legal AI hallucinates in 1 of 6 queries underscores the need for continued human verification.
The Mata v. Avianca incident, while embarrassing for the lawyers involved, served as a useful catalyst for the profession. Courts and bar associations responded quickly with rules and guidance, and the legal profession's self-regulatory apparatus has adapted faster than many other industries to the challenges posed by AI. Over 30 states now have AI-specific guidance for attorneys.
Access to justice remains the area where AI could have the most positive social impact, but it is also the area most at risk of being underserved. The most capable AI legal tools are expensive, and the organizations that serve low-income populations often lack the budgets to adopt them. Whether AI narrows or widens the justice gap will depend significantly on pricing decisions by technology vendors and policy choices by bar associations and courts.
Looking ahead to the rest of 2026, the industry expects agentic AI (systems that can autonomously execute multi-step legal tasks) to move from demonstration to production. Thomson Reuters' agentic CoCounsel features and Harvey's Workflow Builder are early examples of this shift. The question is no longer whether AI will transform legal practice, but how quickly, and who will benefit most from the transformation.