The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.
A daily newsletter distilling the past two weeks of movement in a domain or two — delivered to your inbox while the index updates in the background.
Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail
AI that performs legal research, analyses case law, and assists with e-discovery document review and classification. Includes case law semantic search and predictive coding for document review; distinct from litigation prediction which forecasts outcomes rather than finding relevant materials.
AI-driven legal research and e-discovery have reached a critical inflection point: adoption is now mandatory organizational infrastructure in BigLaw and in-house legal, yet reliability barriers and liability frameworks are hardening faster than technology is improving. Technology-assisted review earned judicial acceptance over a decade ago, and cloud platforms commodified multimillion-document e-discovery long before generative AI arrived. Generative AI now extends legal research, privilege review, and case strategy capabilities into production workflows across every major vendor, with documented productivity gains of 50–70% on large matters and 30% matter capacity increases in large firms. However, the defining tension has intensified: hallucination rates in specialized legal research tools remain material (17–34% across Lexis+ AI and Westlaw), and courts have shifted liability assessment from user error to tool architecture, questioning whether generatively-trained models are inherently unsuitable for verified citation work. The bifurcation is now stark — GenAI matured for high-verification-tolerance cases (privilege logging, government investigations, document review with multi-pass human audit) but mainstream legal research deployment faces escalating regulatory and judicial friction. Legal AI adoption reached 92% of practicing lawyers as of April 2026 globally, with 80% of GenAI users relying on AI for legal research; in-house counsel adoption jumped to 52% (up from 23% a year prior). Yet only 17.7% of e-discovery professionals deploy GenAI on most cases, signaling gap between user adoption and organizational scale deployment. Verification burden persists even at Am Law 100 firms: Sullivan & Cromwell filed briefs with ~40 fabricated citations despite comprehensive AI policies, and courts now require written certification that citations exist and are accurate. Organisations matching use case to verification tolerance extract value. Those treating AI legal research as a feature rather than a liability surface are facing escalating sanctions ($1K–$86K range) and bar discipline.
Infrastructure-level BigLaw commitment reached critical mass in June 2026. Thomson Reuters CoCounsel occupies 1 million seats across 107 countries (80% Am Law 100), with Forrester case studies documenting 30% matter capacity increases and 82% time savings at large firms. Anthropic entered market May 12 with Claude for Legal (20+ integrations, 12 practice-area plugins), deployed at Freshfields (500% usage growth), Quinn Emanuel, Holland & Knight, and Crosby Legal. Relativity’s aiR suite achieved GA across Review (50+ customers post-September 2024), Case Strategy (50+ customers, January 2026 GA), Privilege (99%+ recall, 70% precision, 20-week-to-2-week compression), and Data Breach Response (240M defensible predictions). Harvey AI’s forward-deployed engineer model (6–9 month firm embedments at Allen & Overy, PwC, Cleary Gottlieb, dozens of AmLaw 100 firms) now serves as de-facto BigLaw deployment template. Everlaw AI Assistant serves 125 organizations with precision/recall (0.77/0.82) surpassing human first-pass review; Purpose Legal achieved 98% validated recall on 51,000-document review with 70% precision and $24K savings. Public-sector scaling accelerated: Minnesota Board of Public Defense deployed Vincent AI across 10 judicial districts serving 950 public defenders and 150,000 cases annually; pilot metrics showed witness interview review time collapsed from hours to under 5 minutes, with 300% active user growth within 9 months. Market consolidation signaled maturity: Casetext ($650M to Thomson Reuters), vLex ($1B acquisition by Clio) with Kirkland & Ellis announcing $500M multi-year AI investment and OpenAI launching dedicated legal vertical. Global adoption acceleration confirmed: 92% of lawyers across US, China, and 9 EU countries use AI tools daily (April 2026); 80% of GenAI-using lawyers rely on AI for legal research; 62% report 6–20% time savings. In-house counsel adoption doubled to 52% in US (from 23% prior year), with 64% expecting reduced outside counsel reliance. Deep-integration firms (27% of UK/Ireland sample) report 81% faster client response, 77% improved legal output quality, and 71% cost reduction per matter.
Judicial treatment of generative AI in e-discovery evolved in July 2026: Northern District of California (Schulte v. LinkedIn, June 30) validated LinkedIn's production use of Relativity aiR for document review, applying existing Technology Assisted Review (TAR) principles without requiring special governance or heightened disclosure standards. The court's decision treating generative AI review as an extension of established TAR practice—not a novel legal category—reduces deployment friction for organizations and accelerates adoption of validated tools. Reliability barriers and regulatory friction are hardening in parallel, however. Courts have shifted liability focus from user error to tool architecture: Q1 2026 sanctions ($145K across ~12 cases) now question whether generatively-trained models are architecturally suitable for verified citation work. Ninth Circuit (June 2026, Lnu v. Blanche) sanctioned two attorneys with suspension and $2,500 fines for hallucinated cases and false quotations, establishing attorney duty to verify as non-delegable—attorney gatekeeping responsibility cannot be delegated to tool vendors. Hallucination rates in specialized tools remain material: Princeton’s LePhantomCite benchmark (May 2026) finds 6.57% rate in GPT-5.1 but 17% (Lexis+ AI) and 34% (Westlaw) in domain-trained tools; Stanford research documents 17–33% hallucination rates on representative queries. Florida Supreme Court issued rule amendments (effective June 15, 2026) requiring certification that cited cases exist and are accurate, with sanctions authority—represents systematic regulatory response beyond case-by-case enforcement. Hallucination case tracker documents 1,031+ cases globally with sanctions $1K–$86K and monthly acceleration of 30–50 new cases. Sullivan & Cromwell (Am Law 100, April 2026) filed brief with ~40 AI-fabricated citations despite comprehensive policies and training, demonstrating verification burden persists even at elite tier. EU AI Act Article 50 (effective August 2026) classifies AI document review systems as high-risk with fines up to EUR 35M or 7% global revenue. Governance gaps remain acute: only 17.7% of e-discovery professionals deploy GenAI on most cases; 96% of in-house teams adopted AI but only 31% at scale (67% stuck in pilots); 80% of in-house counsel not requiring outside counsel to use AI and 59% seeing no cost savings from firms’ deployments, revealing misalignment between in-house adoption and external counsel incentive structures. German Bar Association (June 2026, 1,700 lawyers) reported adoption acceleration to 69% YoY but governance crisis: only 9% of legal professionals have enforced AI policies; 43% have no policy at all; 54% receive zero training. The practice remains sharply bifurcated: GenAI proven for high-verification-tolerance cases (privilege, government investigations, document review with extensive multi-pass human audit) but mainstream legal research and document review face unresolved friction from hallucination rates, verification burden, liability shift to tool architecture, and evolving regulatory compliance costs.
— Global survey of 850 senior legal professionals: only 33% trust AI-assisted legal work; 67% concerned verification costs outweigh productivity benefits; 48% report humans always/often materially rewrite AI outputs; only 26% confident their organization's data is AI-ready—documents structural trust and verification barriers limiting enterprise adoption.
— Thomson Reuters earnings call disclosed CoCounsel monthly users quadrupled YoY, Westlaw deep-research searches up 7x in 6 months; 30% of ACV from GenAI products (up from 15% five quarters prior); law firm AI-revenue growth reached 11% through 8 months.
— Thomson Reuters disclosed CoCounsel reached 1 million users across 107 countries, with next-generation product shifting to fully agentic infrastructure; beta research involved 500+ legal professionals across 14 practice areas identifying 40+ legal processes as AI-automation opportunities.
— Independent benchmarking study across major legal AI platforms quantified hallucination rates: 8-17% on legal research tasks, 6-13% on contract review, rising to 23% on state administrative/tribal court matters; demonstrates material error concentrations constraining adoption in high-stakes work.
— Redgrave LLP conducted head-to-head empirical comparison on 45,000-document collection: Relativity aiR achieved 87% recall vs Active Learning's 64% recall; tested against Sedona TAR Reference Model with academic-style rigor, demonstrating measurable precision/recall improvements in production review workflows.
— RelativityOne July 2026 release: custom analyses now support vision models for JPEG/PNG/GIF images; aiR Assist (conversational AI grounded in workspace data) achieved GA; represents incremental maturity in deployed AI review tooling and vision-model expansion.
— EDRM/ComplexDiscovery benchmark (49 respondents, April–May 2026): 69.39% generative AI deployment across e-discovery orgs (4th consecutive increase); 57.14% document controls but 29.41% of deployers run production AI with absent/inconsistent rules—sustained high adoption with widening governance maturity gap.
— Northern District of California (June 30, 2026) approved LinkedIn's production use of Relativity aiR; court applied existing TAR framework rather than creating novel AI rules, rejecting all plaintiff challenges—judicial validation of GenAI in e-discovery with precedent-setting implications.