The State of Play

AI-assisted litigation support has reached a paradox: the tooling works, but the profession's liability framework prevents it from scaling. Forward-leaning firms report dramatic efficiency gains -- 30-40% reductions in drafting time, discovery timelines compressed from months to weeks -- and vendor platforms have matured into production-grade products with million-plus user bases. Yet most of the legal profession has not adopted these tools, and the reason is structural rather than technical. Hallucination rates remain high enough (independent research documents 17-33% error rates on leading platforms) that courts continue sanctioning attorneys who file AI-drafted briefs without adequate verification. Non-delegable attorney accountability means every AI-generated paragraph requires human review, creating a governance and training burden that only well-resourced firms can sustain. The practice sits firmly at leading-edge: a vanguard of large firms is extracting real value, while the broader market watches from the sidelines, blocked not by scepticism about AI's potential but by the unresolved tension between productivity gains and professional liability.

By June 2026, vendor platforms have achieved deployment scale while legal infrastructure gaps persist. Thomson Reuters' CoCounsel serves 1+ million professionals across 107 countries with new litigation-specific features: Litigation Document Analyzer generating counterarguments and supporting case law (Canada, June 2026), and Internal Investigation workflow for defensible incident assessment. Large firms show sustained production adoption: independent Forrester study (June 2026) documented composite 500-attorney firm absorbing 31% more matter volume with 82% time savings, 76% quality improvement, and 53% improved client retention. Quinn Emanuel (1,100-lawyer litigation boutique) operationalized three AI deployment layers: conversational matter intake compressing 90-minute client intake plus 12 associate hours into structured chronology; AI document review across multi-terabyte productions; and adversarial deposition preparation with agent-driven witness simulation. Deposition-specific tools reached scale adoption: Littler (1,000+ attorneys) deployed DepoSim firmwide by May 2026, with the simulator achieving 97% pilot approval across six major firms. Mid-sized firm adoption accelerated to 63% with 75%+ of large firms deploying production AI tools for litigation.

Yet structural adoption barriers remain unresolved. Axiom's June 2026 survey of 500+ corporate legal leaders reveals the pilot-to-production ceiling: 96% adoption breadth but only 31% deployed at scale, two-thirds stuck in pilots. Major barriers: 49% cite bewildering provider fragmentation, 44% data security, 43% accuracy/reliability concerns, 35% internal skill gaps. Data governance emerged as a critical prerequisite for AI effectiveness—most firms' litigation data fragmented across shared drives, emails, billing systems, creating unreliable AI inputs that offset tool sophistication. Hallucination risks persist: peer-reviewed research (Magesh et al., Journal of Empirical Legal Studies 2025) shows Westlaw AI hallucinating 17% and Lexis AI 33% on representative legal queries—benchmarks cited in Ninth Circuit appellate sanctions (June 3, 2026). Global documented hallucination cases reached 1,227+ with Q1 2026 sanctions totaling $145,000. The structural divide has hardened: enterprise adopters with institutional governance infrastructure implementing verification protocols and extracting documented ROI, versus 43% of firms lacking infrastructure for responsible deployment entirely. Courts continue establishing nondelegable verification duties with career-ending penalties for failures (six-month suspensions, bar referrals, $100K+ sanctions imposed in April-May 2026 alone), while 61.6% of federal judges use AI without disclosure or training requirements—a structural asymmetry that persists.

• 2023-H1: Major law firm deployments began (Fisher Phillips, 500+ attorneys) and AI drafting tools matured (Clearbrief-Fastcase integration). However, ChatGPT-generated fake citations in Mata v. Avianca briefs triggered judicial sanctions and standing orders for AI disclosure, establishing the core adoption barrier: verification burden and liability risk for courts and firms.

• 2023-H2: Vendor ecosystem integrated further (Clearbrief-MyCase partnership, advanced features like hyperlinked timelines and document querying). However, hallucination crisis persisted with 10+ additional documented cases of AI-fabricated citations filed in court. Global regulatory response consolidated: courts across jurisdictions (U.S., Canada, UK, DIFC) issued formal guidance requiring AI disclosure, verification, and attorney liability. Adoption survey showed 40% intent (12% active, 28% planning) but 72% industry concern about readiness. Verification burden emerged as the dominant constraint on adoption scaling.

• 2024-Q1: Thomson Reuters (post-Casetext acquisition) expanded CoCounsel Core to Canada and Australia, demonstrating international rollout confidence. However, Stanford research (January 2024) confirmed hallucination crisis remained acute—69-88% error rates on legal reasoning tasks. Sanctions precedent continued with $2,000 February 2024 sanctions for AI-fabricated case citations. Survey data showed broadening organizational interest (60% of attorneys tried AI, 75% of general counsel expect deployment), but Deloitte noted many firms stuck in proof-of-concept, lacking scaling strategies. Verification burden and error rates kept adoption primarily within early-adopter segment.

• 2024-Q2: Thomson Reuters closed acquisition of Casetext ($650M) and announced CoCounsel portfolio expansion across legal, tax, and risk products, broadening vendor integration. Specialist vendors launched new capabilities: U.S. Legal Support released DepoSummary Pro for AI-generated deposition summaries, expanding adoption in litigation support infrastructure. Practitioner adoption signals emerged: law firm interviews reported active use of AI for witness statement consistency checking as part of deposition preparation, indicating adoption moving beyond proof-of-concept. However, K&L Gates and other firm analyses noted that growth in adoption created parallel growth in litigation risk exposure, emphasizing governance and contractual protections as prerequisites for scaled deployment.

• 2024-Q3: Major vendors advanced litigation support capabilities: Thomson Reuters launched CoCounsel 2.0 with 3x performance gains and Claims Explorer for deposition prep (August), while LexisNexis enhanced Lexis+ AI with Claude 3, GPT-4o, and Shepard's GraphRAG integration (July). Deposition-specific tools proliferated: Filevine launched Depo CoPilot for real-time analysis and goal tracking (September). Survey data showed broad organizational interest (85% of law firms/in-house legal see GenAI applicability) but stark market segmentation (76% of largest firms using AI in eDiscovery vs. 28% of smallest). However, adoption barriers remained unchanged: Stanford hallucination research showed no improvement (69-88% error rates); bar associations formalized AI governance (NHBA ethics opinion September 2024) emphasizing attorney verification duty; and sanctions precedent established clear attorney liability for AI-fabricated content. Mainstream adoption remained constrained by verification burden and litigation risk exposure.

• 2024-Q4: Vendor platform maturation continued: Thomson Reuters deepened CoCounsel 2.0 integration with accuracy enhancements (mischaracterization identification, jurisdictional surveys) and launched AI for Justice Legal Aid access program (October). Deposition-specific deployment expanded: Filevine's Depo CoPilot gained traction as real-time analysis tool in production settings (November). Industry adoption metrics showed persistence: 76% of corporate legal departments and 68% of law firms use GenAI weekly (Wolters Kluwer, October), with briefs/memos as top use cases. However, hallucination barrier remained fundamental and unresolved: new sanctions case in Colorado (December) extended liability precedent, and law firm analyses (Baker Botts, December) documented that hallucination risks persisted unchanged. Bar associations and courts reinforced verification requirements as non-delegable attorney duty. By year-end 2024, the practice had achieved managed early-adopter deployment with vendor platform advancement and deposition-specific tools, but mainstream adoption remained blocked by unimproved error rates, liability exposure, and verification burden.

• 2025-Q1: Vendor international expansion accelerated: Thomson Reuters launched next-gen CoCounsel in five new markets (Australia/NZ, Hong Kong, Japan, Southeast Asia, UAE) with 3x faster performance and deeper integrations (March 2025), signaling sustained confidence in platform scaling. Practitioner adoption deepened: Morris Law announced firmwide CoCounsel adoption after 1.5-year evaluation, and American Arbitration Association piloted Clearbrief with arbitrators saving 8-10 hours per case, demonstrating real-world efficiency gains in dispute resolution document work. Adoption metrics confirmed growth trajectory: ABA Tech Survey showed AI adoption tripled to 30% in law firms with CoCounsel at 26% of respondents; CLOC reported 54% of corporate legal departments planning adoption within 2 years. However, hallucination and liability barriers showed no improvement: federal courts issued new sanctions in Nguyen v. Wheeler (E.D. Ark.) and Bunce v. Visual Tech (E.D. Pa.) for AI-fabricated citations, extending 2025 sanctions precedent and reaffirming that verification remains a non-delegable attorney duty. By Q1 2025, the practice had transitioned to sustained mainstream adoption with expanding enterprise customer base and international platform rollout, but acceleration to scaled adoption remained constrained by persistent hallucination risks, court-established liability precedent, and the mandatory verification burden that defines the practice's risk-benefit boundary.

• 2025-Q2: Practitioner adoption continued expanding: UK firm Primas Law (60 staff) deployed CoCounsel across multiple practice areas including litigation, reporting efficiency gains and competitive advantage within six months; three US law firms (Parker Taylor, Drake, Milon) documented specific productivity gains (5-8 hours per deposition, 80% draft completion on summary judgment oppositions). Product ecosystem matured: Clearbrief released advanced cite-checking and timeline-generation capabilities in Word integration, expanding specialized tools for deposition and brief support. However, the hallucination and liability barrier intensified visibly: independent Yale-published peer-reviewed research (Matt Dahl, Journal of Empirical Legal Studies) found Lexis+ AI and Westlaw AI hallucinate 17-33% of the time, challenging vendor marketing claims; multiple 2025 federal sanctions confirmed the barrier remained acute with K&L Gates ($31,100 for nine false citations), Latham & Watkins, and Butler Snow all penalized for AI-generated fabrications; aggregate data showed 129 global hallucination cases in court filings (91 US) with 59 lawyer-sourced, averaging $4,713 per fine and accelerating (32 cases in May alone). Real-time deposition AI tool adoption (Verbit Legal Visor, Deposely) raised ethical concerns about unauthorized practice of law and confidentiality risks, with early deployments by firms like Fisher Phillips highlighted as practice boundaries. By Q2 2025, the practice remained constrained by persistent and intensifying verification burden and hallucination liability risks despite expanded practitioner adoption and vendor platform maturation.

• 2025-Q3: Vendor product ecosystem consolidated: Clearbrief-LexisNexis direct citation-verification integration (September) and Reveal-Clearbrief discovery-to-drafting workflow (August) enabled specialized litigation support deployment. Real-world adoption documented: CoCounsel reached 45+ large firms (50,000+ lawyers) with named cases like Bassford Remele (6-8 hours saved daily), Fisher Phillips (5-hour research to 5 minutes), and Clearbrief achieving 20% cost reduction in appellate work at Falcon Law and adoption by Microsoft legal teams. Survey data confirmed adoption persistence: 59% of legal professionals use AI for briefs/memos, 53% report organizational ROI, 80% expect high impact. However, hallucination crisis intensified acutely: Thomson Reuters identified 22 AI-fabricated citation cases in 33 days (June 30—Aug 1, 2025); novel Noland v. Land of the Free case (California, Sept 2025) sanctioned attorney $10,000 for 21 fabricated citations and raised new liability question—whether opposing counsel has duty to detect opponent's hallucinations. Critically, adoption barrier assessment revealed structural failure: Tillion.ai documented 95% of AI pilots fail to deliver measurable ROI, with 38% of in-house teams actively using AI, widespread "pilot purgatory," and governance gaps preventing transition from proof-of-concept to production. By Q3 2025, the practice remained defined by a dual pattern: enterprise early-adopter deployment under structured verification protocols with documented efficiency gains, versus broader organizational scaling blocked by intensifying error evidence, expanding legal liability precedent, governance and ROI quantification gaps, and the mandatory attorney accountability framework.

• 2025-Q4: Vendor platform consolidation continued with Thomson Reuters launching beta Deep Research on Practical Law for agentic litigation research and deepening CoCounsel-HighQ integration (October, $200M annual AI investment); LexisNexis released Protégé General AI with "Best Fit" multi-model selection (Claude Sonnet 4.5, GPT-5.1) and Shepard's Citation Agent for verification-assisted workflows (December). Adoption metrics showed continued expansion: ACC survey (October) documented corporate legal GenAI adoption doubling to 52% (from 23% in 2024) with 91% reporting efficiency gains in drafting/research, though 60% reported no measurable savings—highlighting persistent ROI quantification gap. However, hallucination and liability barriers showed no structural improvement and intensified in visibility: October 2025 Bloomberg Law analysis found 66 court opinions sanctioning AI misuse and 232 local rules addressing AI use, with courts "growing less lenient" on penalties; Canadian law firm analysis (Osler, December) documented advantages (data review, drafting efficiency) alongside persistent perils (accuracy concerns, hallucinated citations, professional responsibility exposure). By Q4 2025, the practice remained at the leading-edge tier boundary: vendor platforms advancing with agentic capabilities and verification tools, early-adopter deployment continuing with documented productivity benefits, yet mainstream scaling remained structurally blocked by unresolved hallucination risks, expanding legal liability precedent, ROI measurement gaps, and court-mandated verification duties defining the practice's permanent constraint.

• 2026-Jan: Vendor platform expansion accelerated with Thomson Reuters expanding CoCounsel Legal to UK (January 26) featuring deep research on Westlaw/Practical Law and tabular analysis, and LexisNexis launching commercial preview of Protégé AI Workflows (January 21) with hundreds of pre-built litigation tools (motions, discovery, depositions). Practitioner adoption signals continued: Dykema law firm partnership with Clearbrief integrated AI drafting into Microsoft Word. However, hallucination and liability barriers intensified internationally: Abu Dhabi Global Market Court sanctioned law firm AED 282,508 for defence with fictitious cases due to AI misuse (January 12), and New York appellate court sanctioned attorney $5,000 for 23 AI-hallucinated citations in foreclosure brief (January 28). Field reporting documented persistent verification burden: Maryland law firms achieved 30-40% time reduction in writing/research and discovery acceleration from 9 months to 2 weeks, but hallucination rates remained ~1 in 10, forcing continued mandatory human verification. By month-end, the practice showed sustained vendor platform investment and practitioner deployment momentum, but international expansion of hallucination liability and persistent ~10% error rates reinforced structural barriers to mainstream scaling.

• 2026-Feb: Vendor platform maturation reached critical scale: CoCounsel achieved 1 million users across 107 countries (February 26), confirming transition from pilots to production deployment, and LexisNexis released Lexis+ with Protégé GA (February 24) offering 300+ pre-built workflows with Shepard's verification integration. Deposition-specific deployment expanded: DepoSim AI simulator validated in pilot with six major firms (Orrick, K&L Gates, McDermott, Littler, Taft, Brownstein Hyatt), achieving 97% approval and strong reuse intent. Adoption metrics confirmed growth: 75%+ of large law firms running production AI tools for litigation with 70-90% cost reduction in document review. However, fundamental barriers remained unchanged: courts continued sanctioning attorneys for hallucinated citations in 2026 (Lacey v. State Farm, Mid Central Health Fund cases); industry analysis (Thomson Reuters white paper) documented that 80% of legal AI investments fail to deliver measurable ROI despite 75% of legal professionals expecting high impact; post-pilot scaling required infrastructure investment (governance protocols, workflow integration, data handling) that many organizations lacked. By month-end, the practice showed full ecosystem maturation (1M+ user platforms, GA products, specialized deposition tools) but remained constrained by unresolved verification burdens, ROI quantification failures, and scaling barriers that prevented transition to mainstream adoption beyond leading-edge early-adopter segment.

• 2026-Apr: The hallucination crisis reached a quantified milestone: 1,227 documented global cases, Q1 2026 sanctions at a record $145K including a $109,700 Oregon penalty, and the Sixth Circuit (U.S. v. Farris, Apr 3) imposing denial of compensation and disciplinary referral for CoCounsel-fabricated quotations in appellate briefs. Simultaneously, Taft Law (Am Law 100, 350+ litigators) completed firmwide rollout of DepoSim—the first large-scale production deployment of a dedicated deposition preparation simulator—demonstrating that deposition-specific AI can achieve enterprise adoption where brief-drafting tools remain constrained by liability. By late April, the enforcement pattern had intensified: Nebraska Supreme Court indefinitely suspended attorney Greg Lake (April 16) for filing divorce appeal with 57/63 defective citations (90.5% error rate), Third Circuit sanctioned attorney in McCarthy v. DEA for AI-generated fake case summaries (March 27), Puerto Rico Supreme Court issued first monetary penalty in jurisdiction ($1,000, April 27) for false citations with warning of suspension for repeat conduct, and North District of Illinois imposed $9,750 sanction in Chinedu Obi case for 13 fabricated cases and quotations. The Oregon Ghiorso case ($109,700 aggregate—the largest single-attorney AI sanction to date) involved 15 fabricated cases and 9 invented quotations in a filed brief, with post-mortem analysis identifying next-token prediction without source verification as the root cause and retrieval-grounded citation as the proposed architectural remedy. EU regulatory framework codified the constraints: EU AI Act (August 2, 2026 enforcement) explicitly classified litigation support tools as high-risk with specific compliance requirements and penalties. Citation verification solution vendors (BriefCatch RealityCheck) advanced detection engines with demonstrated accuracy catching errors that courts had missed. However, the persistent enforcement pattern—escalating penalties ($15K+ per attorney in 6th Circuit, $109,700 aggregate in Oregon) alongside persistent hallucination rates of 17-34% in benchmark testing—reinforced that liability remains the primary scaling barrier, not tool capability. The practice's dual-tier structure solidified: deposition-specific tools achieving mainstream adoption (DepoSim 97% pilot approval), brief-drafting tools remaining constrained to early-adopters with governance infrastructure.

• 2026-May: Enforcement intensified further with five new federal sanctions emerging in a single 28-day window (May 2-26). S.D.N.Y. sanctioned attorney Tricia Lindsay $2,500 for fabricated citations in litigation memoranda; N.D. California sanctioned Webb Law Group partner Lenden Webb $1,001 for failing to supervise CoCounsel-generated citations in discovery filings, with court explicitly rejecting tool-blame excuses; D.N.J. court documented pro se filer using AI-generated fabricated quotations and non-existent case citations in TCPA opposition brief; C.D. California affirmed $31,100 sanction (Lacey v. State Farm) for multi-firm brief with fabricated citations using CoCounsel, Westlaw Precision Drafting, and Gemini; D. Colorado imposed $3,000 per-attorney sanctions for briefs containing "nearly thirty defective citations" generated by AI legal research tools. Critically, a countervailing technical signal emerged: Digital Applied case study documented a legal research firm achieving 99.2% citation accuracy and 78% weekly adoption using hybrid retrieval RAG—demonstrating that technical solutions to the hallucination barrier exist and can achieve production-grade reliability. Simultaneously, positive deployment signals continued: Littler (1,000+ attorneys) deployed DepoSim firmwide, and Thomson Reuters documented case studies showing Bassford Remele saving 6-8 hours daily, Justly Prudent doubling caseload without hiring, and aggregate ROI of 240 hours annually per attorney (six work weeks). The May 2026 landscape thus revealed the practice's core structural tension: hallucination liability remains the dominant adoption barrier for brief-drafting tools (enforcement escalating despite vendor platform maturity), yet technical solutions exist and deposition-specific tools continue achieving mainstream adoption. The practice remains bifurcated: governance-intensive early adopters with structured verification protocols and emerging technical solutions (RAG, citation agents) deploying successfully in production, versus firms lacking infrastructure continuing to face court sanctions for unverified outputs.

• 2026-Jun: Vendor platform maturation and enterprise adoption momentum reached quantified inflection, but structural adoption ceiling persisted. Thomson Reuters advanced CoCounsel Legal with June releases: Litigation Document Analyzer using generative AI to generate counterarguments and supporting case law for anticipated opposing positions; Internal Investigation workflow automating structured incident assessment and defensible report generation. Independent Forrester economic impact study (published June 8) documented large-firm adoption ROI: composite 500-attorney firm absorbed 31% additional matter volume, 82% time savings, 76% quality improvement, and 53% improved client retention—highest ROI segment across firm sizes, signaling sustained productivity value for well-resourced organizations. Strategic adoption deepened: Quinn Emanuel (1,100-lawyer litigation boutique) operationalized three AI deployment layers—conversational client intake compressing 90 minutes plus 12 associate hours into structured chronology, AI document review at scale, and adversarial deposition preparation with agent-driven witness simulation identifying weak spots and document divergences. Deposition-specific scale achieved: Littler (1,000+ attorneys) completed firmwide DepoSim deployment with 97% pilot approval documented across six major firms (Orrick, K&L Gates, McDermott, Littler, Taft, Brownstein Hyatt). However, the pilot-to-production gap crystallized: Axiom survey of 500+ corporate legal leaders revealed 96% adoption breadth but only 31% at scale deployment, two-thirds stuck in pilots; 49% cite fragmented vendor ecosystem, 44% data security, 43% accuracy/reliability as barriers. Data governance emerged as critical prerequisite: most firms' litigation data fragmented across systems, creating unreliable AI inputs offsetting tool sophistication. Peer-reviewed hallucination benchmarks cited in appellate sanctions (Ninth Circuit, June 3): Westlaw AI 17% error rate, Lexis AI 33%—research published Journal of Empirical Legal Studies 2025 establishing scientific basis for liability. Global institutional response: India Supreme Court released comprehensive draft AI governance framework (June 3-4) explicitly permitting legal research and citation verification with mandatory human oversight while prohibiting algorithmic adjudication, signaling systemic recognition of litigation support as regulated domain. By month-end, the practice demonstrated sustained dual-track pattern: enterprise early adopters with governance infrastructure extracting documented productivity gains (240 hours/attorney/year, 5x ROI, 31% matter volume increases), versus broader profession trapped in pilot purgatory by data governance gaps, hallucination liability, and lack of institutional infrastructure—43% of firms entirely lacking prerequisites for responsible deployment.